Skip to main content
main-content

Über dieses Buch

This book constitutes the refereed proceedings of the 12th International Workshop on Breast Imaging, IWDM 2014, held in Gifu City, Japan, in June/July 2014. The 24 revised full papers and 73 revised poster papers presented together with 6 invited talks were carefully reviewed and selected from 122 submissions. The papers are organized in topical sections on screening outcomes, ultrasound, breast density, imaging physics, CAD, tomosynthesis and ICT and image processing.

Inhaltsverzeichnis

Frontmatter

Invited Talks

Virtual Clinical Trials for the Assessment of Novel Breast Screening Modalities

Validation of any imaging system is challenging due to the huge number of system parameters that should be evaluated. The ultimate metric of system performance is a clinical trial. However, the use of clinical trials is limited by cost and duration. We are strong proponents of a preclinical alternative, in the form of Virtual Clinical Trials (VCT), which model human anatomy, image acquisition, display and processing, and image analysis and interpretation. A complete VCT pipeline was envisioned by combining the breast anatomy and image acquisition simulation pipeline developed at the University of Pennsylvania, with the MeVIC image display and observation pipeline developed by researchers at Barco. Today an integrated virtual clinical trial design program,

VCTdesigner

, and a virtual clinical trial management program,

VCTmanager

, are freely available (www.VCTworld.org). The pipeline design is flexible and extensible, making it possible to add functionality easily and rapidly. It is our hope that by freely distributing the VCTmanager software, our field can standardize on this platform for running VCT.

Andrew D. A. Maidment

Computer-Aided Diagnosis for B-Mode, Elastography and Automated Breast Ultrasound

This review paper encapsulates the presentation of the computer-aided diagnosis (CAD) development in the session of US imaging at IWDM 2014. The development includes novel methodologies in conventional B-mode and modern ultrasound modalities such as elastography and automated breast ultrasound. For B-mode images, gray-scale invariant texture features were proposed to solve the changing of echogenicities from various ultrasound systems. Speckle patterns were analyzed to show the properties of tiny scatterers with microstructure contained in breast tissues for tissue characterization. Using quantified sonographic findings in tumor classification can achieve better diagnostic result than combining all features together. Elastography CAD systems use automatic tumor segmentation and clustering method to reduce operator-dependence. Dynamic sequence features were extracted from a sequence of elastograms to provide tumor stiffness without selecting slices. Another approach was selecting slices with quality evaluation methods. Both approaches reduced the overloads of physicians in slice selection. Automated breast ultrasound system is developed to automatically scan the whole breast and build the volumetric breast structure. Three-dimensional morphology, texture, and speckle features were proposed and combined to provide more diagnostic information than two-dimensional features. These CAD systems for B-mode, elastography, and automated breast ultrasound are good at malignancy evaluation and would be helpful in clinic use.

Ruey-Feng Chang, Chung-Ming Lo

Measurement and Clinical Use of Breast Density

Breast density is loosely defined as the amount of fibroglandular tissue in the breast compared to the total amount of breast tissue. In this review paper we consider the three ways of describing breast density as seen on a mammogram: pattern-based, area-based and volumetric-based and explain the rationale for each along with detailing the various ways of estimating each of them (visual, semi-automated, and fully automated). We also consider the use of other imaging modalities of estimating breast density, including CT. Clinically, breast density has now moved from being a controversial, even derided subject to one which is widely accepted with an expanding number of clinical uses. It is proven that a woman’s breast density is a strong predictor of the failure of mammographic screening to detect breast cancer and thus can be used to indicate where alternate modalities might be considered. It is proven that breast density is a strong predictor of the risk of developing breast cancer and thus can be used to start to consider tailored screening programs. We review the current widely known clinical uses along with the lesser known uses, such as assessing the benefits of chemoprevention and generating more accurate radiation dose estimates. Breast density is becoming an increasingly important clinical tool; there is an increasing need for accurate and consistent density measures along with an understanding of how the various measures compare.

Kwan-Hoong Ng, Susie Lau

Low-Dose Molecular Breast Imaging - Diagnostic and Screening Applications in Women with Dense Breasts

Approaches to imaging the breast with nuclear medicine and/or molecular imaging methods have been under investigation since the early 1990s. Nuclear medicine procedures, which detect the preferential uptake of a radiotracer in breast lesions, have the potential to offer valuable functional information that complements conventional anatomical imaging techniques such as mammography and ultrasound.

Despite initial enthusiasm for scintimammography, nuclear medicine techniques in general have struggled to gain mainstream acceptance by the breast imaging community. In the last 5-10 years, older-generation scintillating gamma systems, such breast-specific gamma imaging systems, have been replaced by a new generation of dual detector cadmium zinc telluride [CZT] detectors that perform direct conversion [DC] of gamma ray energy to signal and yield improved spatial and energy resolution. Using CZT-based detectors, DC-Molecular breast Imaging [DC-MBI] has demonstrated the ability to reliably detect breast tumors in a variety of diagnostic and screening settings. Recent improvements in both the detector technology and patient preparation have enabled the associated radiation dose from DC-MBI to be reduced to less than 1.5 mSv. The most robust evidence for the clinical use of DC-MBI is in the screening of women with dense breast tissue. In two large screening studies, the addition of DC-MBI to mammography was significantly more sensitive than mammography alone in detecting cancer (91% vs. 25%, p <0.001). Supplemental screening with DC-MBI detected an additional 8.3 cancers per 1000 women, which compares very favorably to other modalities in screening women with dense breasts.

Michael K. O’Connor

Will New Technologies Replace Mammography CAD as We Know It?

Since its commercial introduction in 1998, Mammography computer-aided detection (CAD) has been one of the few CAD technologies widely implemented in clinical practice. The original concept of CAD marks as overlays on images has been broadly accepted, although new paradigms have been proposed and successfully tested that could one day challenge that original approach. But now, as breast imaging evolves further with the advent of digital breast tomosynthesis, new image processing techniques are developing that may cause us to re-evaluate the clinical requirements for Mammography CAD as we know it. What clinical problems in breast imaging are not solved by tomosynthesis, and how can CAD help us with those problems?

Julian Marshall, Ashwini Kshirsagar, Sibel Narin, Nikos Gkanatsios

Advanced Telecommunications in Breast Imaging – Streamlining Telemammography, Telepathology and Teleoncology Services to Improve Patient Care

Teleradiology services are very common worldwide, and Full Field Digital Mammography (FFDM) systems have made it possible to include mammography. This is important because breast cancer is the most common cancer in women in many parts of the world and it is the second leading cause of cancer deaths. In rural and medically underserved areas, mammography rates are lower than in urban areas for a variety of reasons, including lack of dedicated screening facilities and/or personnel, poor compliance, and large distances between patients and clinics (making it difficult to return for follow-up care). Once possible cancers are detected, biopsies are performed but in many cases reports are very slow to get back to the patient and/or local clinician impacting treatment and follow-up. Telepathology can address this situation. Patients may also require oncology services and in rural areas they are often limited or non-existent. Real-time teleoncology services can facilitate treatment and counseling. Finally, breast care support and education can be facilitated by virtual support groups and broadcasting education lectures.

Elizabeth A. Krupinski

Screening Outcomes

Predicting the Benefit of Using CADe in Screening Mammography

The goal of computer-aided detection (CADe) in screening mammography is to help radiologist avoid missing breast cancer. Thus when designing a CADe system it is important to know how different methods and parameters would affect radiologists’ ability to use the system effectively. Short of conducting an observer study or a clinical trial, this is not possible. In this paper, we present preliminary results on a model that can predict how many additional cancers a radiologist would detect, if they used a CADe system. The model uses the results of radiologists’ reading of a set of screening mammograms without using CADe to predict the probability that a radiologist would miss a cancer when reading without CADe and the probability that the radiologist would recall the woman if CADe flagged the missed cancer. In our initial study, 8 radiologists read 300 screening mammograms containing 69 cancers with and without CADe. Our model predicted that on average a radiologist would detect 4.7 extra cancers while the actual number of extra cancers detected per radiologist was 3.6. Bootstrapping across readers, the 95% CI for the difference between the predicted and actual number of extra cancers was [-1.52, 3.30]. The overall ability of the model to discriminate between lesions detected as a result of CAD flag and other lesion examinations was moderately high, with c-index of 0.77 (95% CI of [0.6, 0.91]). We are currently conducting a study with larger number of radiologists and cases to obtain better estimates of the accuracy of our model.

Robert M. Nishikawa, Andriy Bandos

Modeling Breast Cancer Screening Outcomes

There is currently rapid development of imaging technologies for breast cancer screening. In addition there is considerable controversy regarding the optimal screening strategy, including the ages at which screening should begin and end, the interval between screens and the imaging modality or modalities which should be used. Furthermore, there are major economic considerations related to whether screening should be done and how it should be done. Here, we describe the use of the Wisconsin CISNET computer model of breast cancer development to predict key outcomes associated with breast cancer, including incidence, mortality and life-years lost due to breast cancer. The sensitivity and specificity of the detection method and their dependence on factors such as age and breast density are implemented in the model through use of empirical data. Distributions of cancer characteristics are used to determine the type of modern therapy utilized and its effectiveness. Using this framework, the effectiveness of a particular screening strategy can be compared with other scenarios such as not screening at all or following published recommendations. The model can directly inform a cost-effectiveness or cost-utility analysis.

Martin J. Yaffe, Nicole Mittman, Natasha Stout, Pablo Lee, Anna Tosteson

The Impact of Introducing Full Field Digital Mammography into a Screening Programme

As digital mammography has largely replaced analogue mammography for screening, we examine the impact of the change on clinical outcomes. Data were obtained for 62,599 women undergoing routine screening mammography between January 2010 and March 2012 during the transition from analogue to digital screening, and on a monthly basis for the two years following complete conversion to digital mammography. With digital mammography, the recall rate increased from 3.58% to 4.69% (p<0.001) and the biopsy/cytology referral rate increased from 1.49% to 1.88% (p=0.01) whilst the cancer detection rate did not change significantly. The recall rate showed a strong positive correlation with time (r=0.71, p<0.001) and biopsy rate showed a moderate positive correlation (r=0.52 p=0.011). The cancer detection rate showed a moderate negative correlation with time (r=-0.567 p=0.05). Whilst all the outcome measures reported meet the standards set by the national screening programme, these results indicate a need for regular monitoring following changes to screening technology.

Thomas Fyall, Caroline Boggis, Jamie Sergeant, Elaine Harkness, Sigrid Whiteside, Julie Morris, Mary Wilson, Susan M. Astley

Ultrasound

Fully Automated Nipple Detection in 3D Breast Ultrasound Images

Nipple position provides useful diagnostic informations in reading automated 3D breast ultrasound (ABUS) images. The identification of nipples is required to localize and determine the quadrants of breast lesions. Additionally, the nipple position serves as an effective landmark to register an ABUS image to other imaging modalities, such as digital mammography, breast magnet resonance imaging (MRI), or tomosynthesis. Nevertheless, the presence of speckle noise induced by interference waves and variant imaging directions in ultrasonography poses challenges to the task. In this work, we propose a fast and automated algorithm to detect nipples in 3D breast ultrasound images. The method fully takes advantages of the consistent characteristics of ultrasonographic signals observed at nipples and employs a multi-scale Laplacian-based blob detector to eventually identify nipple positions. The accuracy of the proposed method was tested on 113 ABUS images, resulting in a distance error of 6.6±8.9 mm (

mean

±

std

).

Lei Wang, Tobias Böhler, Fabian Zöhrer, Joachim Georgii, Claudia Rauh, Peter A. Fasching, Barbara Brehm, Rüdiger Schulz-Wendtland, Matthias W. Beckmann, Michael Uder, Horst K. Hahn

Breast Imaging with 3D Ultrasound Computer Tomography: Results of a First In-vivo Study in Comparison to MRI Images

Ultrasound Computer Tomography (USCT) is a promising modality for breast imaging. We developed and tested the first full 3D USCT system aimed at in-vivo imaging. It is based on approx. 2000 ultrasound transducers surrounding the breast within a water bath. From the acquired signal data, reflectivity, attenuation and sound speed images are reconstructed. In a first in-vivo study we imaged ten patients and compared them to MRI images. To overcome the considerably different breast positioning in both imaging methods, an image registration and image fusion based on biomechanical modeling of the buoyancy effect and surface-based refinement was applied. The resulting images are promising: compared with the MRI ground truth, similar tissue structures can be identified. While reflection images seem to image even small structures, sound speed imaging seems to be the best modality for detecting cancer. The registration of both imaging methods allows browsing the volume images side by side and enables recognition of correlating tissue structures. The first in-vivo study was successfully completed and encourages for a second in-vivo study with a considerably larger number of patients, which is currently ongoing.

Torsten Hopp, Lukas Šroba, Michael Zapf, Robin Dapp, Ernst Kretzek, Hartmut Gemmeke, Nicole V. Ruiter

Breast Density

Factors Affecting Agreement between Breast Density Assessment Using Volumetric Methods and Visual Analogue Scales

Mammographic density in digital mammograms can be assessed visually or using automated volumetric methods; the aim in both cases is to identify women at greater risk of developing breast cancer, and those for whom mammography is less sensitive. Ideally all methods should identify the same women as having high density, but this is not the case in practice. 6422 women were ranked from the highest to lowest density by three methods: Quantra

TM

, Volpara

TM

and visual assessment recorded on Visual Analogue Scales. For each pair of methods the 20 cases with the greatest agreement in rank were compared with the 20 with the least agreement. The presence of microcalcifications, skin folds, suboptimally positioned inframammary folds, and whether or not the nipple was in profile were found to affect agreement between methods (p<0.05). Careful positioning during mammographic imaging should reduce discrepancy, but a greater understanding of the relationship between methods is also required.

Lucy Beattie, Elaine Harkness, Megan Bydder, Jamie Sergeant, Anthony Maxwell, Nicky Barr, Ursula Beetles, Caroline Boggis, Sara Bundred, Soujanya Gadde, Emma Hurley, Anil Jain, Elizabeth Lord, Valerie Reece, Mary Wilson, Paula Stavrinos, D. Gareth Evans, Tony Howell, Susan Astley

Breast Tissue Segmentation and Mammographic Risk Scoring Using Deep Learning

Mammographic scoring of density and texture are established methods to relate to the risk of breast cancer. We present a method that learns descriptive features from unlabeled mammograms and, using these learned features as the input to a simple classifier, address the following tasks: i) breast tissue segmentation ii) scoring of percentage mammographic density (PMD), and iii) scoring of mammographic texture (MT). Our results suggest that the learned PMD scores correlate well to manual ones, and that the learned MT scores are more related to future cancer risk than both manual and automatic PMD scores.

Kersten Petersen, Mads Nielsen, Pengfei Diao, Nico Karssemeijer, Martin Lillholm

Imaging Physics I

Optimization of X-Ray Spectra for Dual-Energy Contrast-Enhanced Breast Imaging: Dependency on CsI Detector Scintillator Thickness

Columnar structured cesium iodide (CsI) scintillators doped with Thallium (Tl) have been used extensively for indirect X-ray imaging detectors. Here, theoretical modeling was performed to assess the impact of CsI thickness on optimal acquisition spectra for dual-energy iodine-enhanced breast computed tomography (bCT). Contrast-to-noise ratio (CNR) between iodine-enhanced and non-enhanced breast tissue normalized to the square root of the total average glandular dose (AGD) was computed as a function of the fraction of the AGD allocated to the low-energy images. Peak CNR/√AGD and optimal low-energy AGD allocations were identified for small, average and large uncompressed breasts. Optimal high-energy spectra were found to be almost independent of CsI thickness and occurred just above the Cs and I K-edges (range 34 to 36 keV), while optimal low-energy spectra varied largely with CsI thickness, ranging from 25 keV to 33 keV for 100

μ

m to infinite CsI scintillator thicknesses.

Pablo Milioni de Carvalho, Ann-Katherine Carton, Sylvie Saab-Puong, Răzvan Iordache, Serge Muller

Dose-Saving Potential of Linear- and Non-Linear Energy Weighting in Photon-Counting Spectral Mammography

Energy weighting techniques are known to improve the contrast-to-noise (CNR) ratio in energy-sensitive, x-ray photon detection, in particular in the absence of scattered radiation. In spite of the rather moderate reported improvements in CNR, typically ranging between 5-10%, it is of high relevance to quantify the potential for saving radiation dose in a mammography screening environment. In this paper we experimentally investigate the possible improvements to be obtained by energy-weighting of data acquired with a Philips MicroDose SI mammography system. We compare three schemes to combine the raw data consisting of counts registered in the low- and high-energy bins, respectively: conventional summation, linear weighting and non-linear weighting of the two energy bins. Measurements on a dedicated phantom were analyzed to quantify the potential for reduction of patient dose of linear and non-linear energy weighting. By averaging improvements of CNR achieved over several pairs of regions-of-interest (ROI) we report a potential to reduce the patient dose by 7% for linear- and 9% for non-linear energy weighting, in good agreement with expectation.

Udo van Stevendaal, Hanno Homann, Ewald Roessl, Klaus Erhard, Björn Cederström

Compositional Three-Component Breast Imaging of Fibroadenoma and Invasive Cancer Lesions: Pilot Study

Purpose: To investigate the lesion discrimination ability of compositional 3-component breast imaging technique (3CB) of patients with suspicious breast lesions (BIRADS 4 or greater).

Materials and Methods: A novel dual-energy 3CB imaging technique concludes in quantifying of the lipid, protein, and water thicknesses. The protocol was designed to be performed on a standard full-field digital mammography system by imaging additional high-energy image using a 3-mm Al filter. A pilot study of 43 abnormal breast findings on diagnostic mammography was performed using the 3CB protocol. The lesion groups include fibroadenoma (FA), invasive (IDC), DCIS and benign tissues. The lesions were delineated by the radiologist on CC and MLO views, and the compositional measures of the whole breasts, local areas within lesions and their peripheries were derived. Univariate logistic regression statistics was applied to analyze lesion different group separation. The variable statistical significance of MLO, CC views and their average was also compared.

Results: We found for FA/rest group discrimination that water and lipid difference between lesion and periphery are significant for CC and MLO views. In addition, the breast fibroglandular dense volume are also significant for both views. Lesion to background water difference predicted FA with an odds ratio = 4.4 , ROC area of 0.8. For cancer/non cancer groups there were no variables showing the significance for both views. However, for IDC/rest groups lipid thicknesses within breast and at the periphery normalized by total thicknesses become significant for both views.

Conclusion: Our pilot set data demonstrates that the technique provides biologically meaningful compositional components of lesion, its periphery and breast which are statistically significant for FA/rest and invasive cancers/rest group separation.

Serghei Malkov, Fred Duewer, Karla Kerlikowske, Karen Drukker, Maryellen Giger, John Shepherd

CAD

Potential Usefulness of Presentation of Histological Classifications with Computer-Aided Diagnosis (CAD) Scheme in Differential Diagnosis of Clustered Microcalcifications on Mammograms

We compared the usefulness of the presentation of likelihood of histological classifications with that of malignancy evaluated by a CAD scheme in the differential diagnosis of clustered microcalcifications (MCs) on magnified spot mammograms. The likelihood of histological classifications was evaluated by the CAD scheme using 5 objective features that radiologists commonly use for describing MCs. The likelihood of malignancy was evaluated based on the likelihood of histological classifications. Unknown cases for an observer study consisted of 22 benign MCs (15 micro-cysts and 7 mastopathies) and 26 malignant MCs (10 DCISs of comedo type and 16 DCISs of noncomedo type). Thirteen observers independently provided their confidence level regarding the malignancy of the unknown case before viewing the evaluated result by the CAD scheme, after viewing the likelihood of malignancy and after viewing the likelihood of histological classifications. The results were evaluated with multi-reader, multi-case receiver operating characteristic (ROC) analysis. The average area under the curve (AUC) for all observers without CAD, with CAD for malignancy and with CAD for histological calcifications was 0.670, 0.802 and 0.819 (

P

< .01), respectively. The presentation of the likelihood of histological classifications improved radiologists’ performance than that of malignancy in the differential diagnosis of MCs.

Ryohei Nakayama, Kiyoshi Namba, Ryoji Watanabe, Hiroshi Nakahara, Ralph Smathers

Potential Usefulness of Breast Radiographers’ Reporting as a Second Opinion for Radiologists’ Reading in Digital Mammography

We investigated a potential usefulness of breast radiographers’ reporting, in terms of a second opinion for improving radiologists’ diagnostic performance in the detection of microcalcifications in digital mammography. This simulation study was conducted, by use of an existing jackknife free-response receiver operating characteristic (JAFROC) observer study data obtained with 75 cases(25 malignant, 25 benign, and 25 normal cases) of digital mammogram, selected form the digital database for screening mammography (DDSM) provided by University of South Florida. Each of rating scores obtained by 6 breast radiographers was utilized as a second opinion for 4 radiologists’ reading with radiographers’ reporting. Average figure of merit (FOM) of radiologists’ performance was generally improved by use of radiographer’s reporting, and significant improvements were found in case of 3 out of 6 radiographers’ reporting used.

Rie Tanaka, Miho Takamori, Yoshikazu Uchiyama, Junji Shiraishi

Tomosynthesis

Effective Detective Quantum Efficiency (eDQE) Measured for a Digital Breast Tomosynthesis System

This paper presents effective detective quantum efficiency (eDQE) results for a digital breast tomosynthesis (DBT) system. Poly(methyl methacrylate) (PMMA) blocks of thickness 2, 4, 6 and 7 cm were imaged under automatic exposure control (AEC) in standard 2D digital (planar) mammography (with anti-scatter grid) and DBT mode (without anti-scatter grid). Modulation transfer function (MTF) for the projection images was measured in the front-back and left-right (i.e. tube-travel for DBT) directions using a 0.8 mm thick steel edge at positions 2, 4, 6 and 7 cm above the breast table. NNPS data required for eDQE calculation were calculated from the AEC projection images (the ~0° projection AEC image for DBT). The eDQE at 0.5 mm

− 1

in planar mammography mode was relatively stable at ~0.25 as PMMA thickness changed from 2 to 7 cm. For DBT, blurring from the focus motion and scattered radiation reduced eDQE at 6 and 7 cm PMMA.

Nicholas Marshall, Elena Salvagnini, Hilde Bosmans

Comparison of SNDR, NPWE Model and Human Observer Results for Spherical Densities and Microcalcifications in Real Patient Backgrounds for 2D Digital Mammography and Breast Tomosynthesis

The development of objective detection performance measures is of great value for both optimization purposes and inter-comparison of different mammographic (2D) or digital breast tomosynthesis (DBT) systems. They would be valuable as an alternative for human observer studies. In this study, we have calculated contrast (C), signal-difference-to-noise ratio (SDNR) and a NPWE model observer (

d’

) for spherical densities and microcalcifications in patient 2D and DBT images. Contrast was higher in 2D compared to DBT reconstructions. In contrary, SDNR values were higher in DBT for the spheres and were comparable between 2D and DBT for the microcalcifications. A comparison between

d’

and SDNR showed a strong correlation between these two measures for both spheres and microcalcifications in 2D and DBT. Only weak and moderate correlations were found for SDNR and d’ versus human confidence scores, indicating that there is room for improvement for the NPWE model observer as a theoretical predictor of human detection performance.

Lesley Cockmartin, Nicholas W. Marshall, Hilde Bosmans

Assessing Radiologist Performance and Microcalcifications Visualization Using Combined 3D Rotating Mammogram (RM) and Digital Breast Tomosynthesis (DBT)

We evaluated the diagnostic performance of a novel 3D visualization approach (rotating mammogram, RM) in combination with DBT compared with that of FFDM and DBT alone.. FFDM, DBT alone and DBT images plus reconstructed RM from 110 breasts (34 cases of breast cancer and 76 normal breasts) were evaluated and rated independently by 6 readers. DBT plus RM demonstrated superior diagnostic accuracy compared to FFDM (p<0.05) and a small improvement in performance compared to DBT alone. Visualization of microcalcifications was significantly better on RM than DBT (p<0.05) for all 14 microcalcification-dominant cancer lesions. Adjunction of RM to DBT will offer the benefit of increased diagnostic accuracy and contribute to more accurate assessment of DBT alone.

Hitomi Tani, Nachiko Uchiyama, Minoru Machida, Mari Kikuchi, Yasuaki Arai, Kyoichi Otsuka, Anna Jerebko, Andreas Fieselmann, Thomas Mertelmeier

Digital Breast Tomosynthesis: Image Quality and Dose Saving of the Synthesized Image

In this paper, the impact on image quality and dose reduction related to the use of a synthesized 2D image in digital breast tomosynthesis examinations is analyzed. 2D and 3D images of the TORMAM phantom were acquired at clinical conditions. Syntesized 2D images (C-View) were also obtained. Seven observers scored the detectability and visibility of microcalcification (MC) clusters in both types of images. Low contrast objects were studied measuring contrast-to-noise ratio (CNR) and applying a non-prewhitening matched filter (NPW) model observer. Glandular doses were estimated from a sample of 50 patients. The detectability and visibility of the microcalcification clusters were higher in C-View than in 2D images (50% and 100%, respectively). CNR values were higher for C-View for all contrasts. The NPW got slightly higher detectability values for the lowest contrast details in C-View. We have estimated a dose reduction of 43% by replacing the conventional 2D by the C-View image.

Julia Garayoa, Irene Hernandez-Giron, Maria Castillo, Julio Valverde, Margarita Chevalier

Patient Specific Dose Calculation Using Volumetric Breast Density for Mammography and Tomosynthesis

Minimising the mean glandular dose (MGD) received by the patient whilst maximising image contrast during mammographic imaging is of paramount importance due to the widespread use of the modality for screening, where subjects are for the most part healthy. The advent of digital mammography brought about a general reduction in MGD, however the introduction of tomosynthesis, particularly when used in combination with conventional projection mammography has the potential for unwanted and often unnecessary MGD increases. We describe a method to calculate the patient-specific MGD using a representation of the patient’s volumetric breast density to derive the breast glandularity. This personalises the MGD to the individual woman, rather than assuming a constant value, or one that depends solely on compressed breast thickness. The calculated patient specific MGDs are compared to those reported by the manufacturer for a database of 2D mammograms. Though agreement is generally good for dense breasts, we have found that the MGD is underestimated in fatty breasts. A separate database of 2D mammogram and 3D tomosynthesis acquisitions acquired in “combo” is also analysed. In general, the MGDs are approximately equal for dense (VDG 3 and 4) breasts, but fatty (VDG 1 and 2) breasts exhibited significant differences with tomosynthesis MGDs being higher than mammogram MGDs for these cases.

Christopher E. Tromans, Ralph Highnam, Oliver Morrish, Richard Black, Lorraine Tucker, Fiona Gilbert, Sir Michael Brady

Imaging Physics II

Comparative Performance Evaluation of Contrast-Detail in Full Field Digital Mammography (FFDM) Systems Using Ideal (Hotelling) Observer versus Automated CDMAM Analysis

The purpose of our work was to evaluate contrast-detail performance for a range of full field digital mammography systems using Hotelling observer SNR analysis and ascertain whether it can be considered an alternative to CDMAM evaluation. Five FFDM systems were evaluated, which differed in generation (age), Automatic Exposure Control (AEC) behaviour, tube/target combination and detector type. Contrast-detail performance was first analysed using CDMAM phantom analysis and then using the Hotelling observer SNR methodology. The Hotelling observer SNR was calculated for input signal originating from gold discs of varying thicknesses and diameters and then used to estimate the threshold gold thicknesses for each diameter as per CDMAM analysis. There were small differences between the two techniques, especially in small diameter details, which can be attributed to structural characteristics of the CDMAM phantom. The Hotelling observer SNR technique showed lower variability than results from CDMAM analysis. Overall, the Hotelling observer SNR methodology showed variations in the FFDM systems performance consistent with previous findings, demonstrating its value as a performance assessment metric.

Ioannis Delakis, Robert Wise, Lauren Morris, Eugenia Kulama

Mammographic Density Effect on Readers’ Performance and Visual Search Pattern

A test set of 150 digital mammograms were examined by 14 radiologists, seven of which underwent eye-position recording. Mammograms were classified into low- and high- density cases, in order to investigate the impact of density on readers’ performance and visual search patterns. Lesions overlaying were compared to those outside the dense fibroglandular tissue. Our results suggest that when the lesion was overlaying the fibroglandular tissue, readers’ performance significantly improved in high- compared to low- density cases. Also the dense area of breast parenchyma attracted the radiologists’ visual attention, in both low- and high- mammographic density cases. When the lesions were outside the dense fibroglandular tissue, no difference was noted in radiologist’ performance. In conclusion, dense areas of the breast parenchyma attracted the radiologists’ visual attention, in both low- and high density cases, which might improve lesion detection when the malignancy is overlaying the dense parts of the breast tissue.

Dana S. AL Mousa, Patrick C. Brennan, Elaine A. Ryan, Claudia Mello-Thoms

Towards a Quantitative Measure of Radiographic Masking by Dense Tissue in Mammography

The detection sensitivity of screening mammography is reduced for dense breasts where the appearance of fibroglandular tissue can mask suspicious lesions. A measure of the degree of masking expected for a mammogram could be useful for informing the decision to direct some women to supplemental imaging procedures not affected by density. Here, we present an adaptation of a model observer to estimate the detection task SNR,

d

local

, of a lesion embedded in various portions of the breast to indicate the level of detection difficulty. Rank correlation of mean mammogram

d

local

with density category is

ρ

=–0.58. Correlation of fractional area of mammograms with low

d

local

 < 2 versus density category is

ρ

=0.61. This suggests that a metric based on

d

local

may be useful in quantifying masking effects of breast density.

James G. Mainprize, Xinying Wang, Mei Ge, Martin J. Yaffe

Three Dimensional Dose Distribution Comparison of Simple and Complex Acquisition Trajectories in Dedicated Breast CT – A Monte Carlo Study

The purpose of this study was to characterize the three dimensional (3D) x-ray dose distributions in a target scanned with different acquisition trajectories for dedicated breast CT imaging. Monte Carlo simulations were used to evaluate two acquisition trajectories: circular azimuthal (no tilt) and complex sinusoidal (saddle) orbit with ±15

o

tilts around a pendant breast. Simulations were performed with tungsten (W) and cerium (Ce) filtration of a W-anode source; the simulated source flux was normalized to the measured exposure of a clinically used W-anode source. A water filled cylindrical phantom, was divided into 1cc voxels, and each voxel was set to track the cumulative energy deposited. Energy deposited per voxel was converted to dose, yielding the 3D distributed dose volumes. Results indicate that the mean absorbed dose at the isocenter of a volume for the un-tilted acquisition is ~10% higher than that from a saddle scan, regardless of filtration used.

Jainil P. Shah, Steve D. Mann, Randolph L. McKinley, Martin P. Tornai

ICT and Image Processing

Quantitative MRI Phenotyping of Breast Cancer across Molecular Classification Subtypes

The goal of our study was to investigate the potential usefulness of quantitative MRI analysis (i.e., phenotyping) in characterizing and data mining the molecular subtypes of breast cancer in order to better understand the difference among HER2, ER, and PR expression, triple negative, and other molecular classifications. Analyses were performed on 168 biopsy-proven breast cancer MRI studies acquired between November 2008 and August 2011, on which molecular classification was known. MRI-based phenotyping analysis included: 3D lesion segmentation based on a fuzzy c-means clustering algorithm, computerized feature extraction, leave-one-out linear stepwise feature selection, and discriminant score estimation using Linear Discriminant Analysis (LDA). The classification performance between the molecular subtypes of breast cancer was evaluated using ROC analysis with area under the ROC curve (AUC) as the figure of merit. AUC values obtained for 26 HER2+ vs. 142 HER2-, 118 ER+ vs. 50 ER-, 93 PR+ vs. 75 PR-, 40 Triple Negative (ER-, PR-, and HER2-) vs. 128 all others are 0.65, 0.70, 0.57, and 0.68, respectively for the combined datasets that included images from both 1.5T and 3T scanners. Contributions to the classifiers come from the shape, texture, and kinetics of the lesion, triple negative cases exhibiting increased margin variability, distinct kinetics, and increased surface area. Analyzing the datasets within magnet strength substantially improved performances, e.g., the AUC for triple negative vs. all other cancer subtypes increased from 0.69 (SE=0.05) to 0.88 (SE=0.05). The results from this study indicate that quantitative MRI analysis shows promise as a means for high-throughput image-based phenotyping in the discrimination of breast cancer subtypes.

Maryellen L. Giger, Hui Li, Li Lan, Hiroyuki Abe, Gillian M. Newstead

A Novel Framework for Fat, Glandular Tissue, Pectoral Muscle and Nipple Segmentation in Full Field Digital Mammograms

Automated segmentation of mammograms is an important initial step in a wide range of applications including breast density and texture analysis and computer aided detection of abnormalities. In this paper, we propose a unified machine learning framework that enables simultaneous segmentation of the breast region, fatty tissue, glandular tissue, pectoral muscle and nipple region in full field digital mammograms. We calculate both a multi-label segmentation mask and a probability map associated with each of the segmented classes. The probability map facilitates interpretation of the segmentation mask prior to further analysis. The method is evaluated using left or right MLO views from 100 women in a 5-fold cross validation manner. Our framework is shown to be robust and accurate, achieving sensitivity/specificity from 82.7% to 98.5% at the equal-error-rate point of the ROC curves and area under the ROC curve values from 0.9220 to 0.9998 for the corresponding segmentations.

Xin Chen, Emmanouil Moschidis, Chris Taylor, Susan Astley

Texture-Based Breast Cancer Prediction in Full-Field Digital Mammograms Using the Dual-Tree Complex Wavelet Transform and Random Forest Classification

In this paper we describe a novel methodology for texture-based breast cancer prediction in full-field digital mammograms. Our method employs the Dual-Tree Complex Wavelet Transform for texture-based image analysis and representation, and Random Forest classification for discriminative learning and breast cancer prediction. We assess the ability of our method to identify women with breast cancer using raw images, processed images and Volpara

TM

density maps of two case-control datasets. We also investigate whether different regions of the breast exhibit different predictive power with respect to breast cancer. The best results are obtained using the processed images of a case-control dataset consisting of 100 cancers and 300 controls, where we achieve an area under the ROC curve of 0.74 for a texture model based on the whole breast and an equal area under the ROC curve when the most predictive regional model is used.

Emmanouil Moschidis, Xin Chen, Chris Taylor, Sue M. Astley

Poster Papers

Evaluation of a New Design of Contrast-Detail Phantom for Mammography: CDMAM Model 4.0

The standard test object used to assess the imaging performance of digital mammography systems in Europe is the CDMAM model 3.4. The recently released CDMAM model 4.0 differs from the model 3.4 in the layout, number and range of thicknesses of gold contrast details used to assess threshold contrast detail detection. In order to evaluate CDMAM 4.0 we compared its performance with that of the CDMAM 3.4 using several digital mammography systems at various dose levels. We also assessed the reproducibility of the results compared to that of the previous model. CDMAM 4.0 results were comparable to results for CDMAM 3.4 for detail diameters in the range 0.1 to 0.5mm and for the larger detail diameters there were increased differences as would be expected due to the design differences of the CDMAM 4.0. The reproducibility of CDMAM 4.0 results was found to be better than that of CDMAM 3.4 results.

Celia J. Strudley, Kenneth C. Young

Threshold Target Thickness Calculated Using a Model Observer as a Quality Control Metric for Digital Mammography

Task-based measures estimated by model observers may provide a more clinically relevant and objective way of assessing image quality and system performance for quality control of digital mammography. One approach is to calculate the required threshold thickness (

t

t

) of a material necessary to render test objects just visible, using the detectability index (

) calculated from measured system parameters and a non-prewhitening observer model incorporating an eye-filter and internal noise (NPWE). Our previous work developed methodology for simply measuring the parameters required to calculate

and

t

t

using a NPWE model. Here we test the sensitivity of

t

t

to changes in image quality by varying entrance exposure and by imaging with and without a grid. Calculated

t

t

values are compared with those reported by automated analysis of CDMAM (

TM

) phantom images (CDCOM). Sensitivity to dose changes was seen, and good correlation was achieved between CDCOM and our model.

Aili K. Bloomquist, James G. Mainprize, Melissa Hill, Martin J. Yaffe

Contrast-Enhanced Digital Mammography Lesion Morphology and a Phantom for Performance Evaluation

Contrast-enhanced digital mammography (CEDM), promises to improve diagnostic accuracy as an adjunct to mammography, especially for women with dense breasts. Here we review 98 enhancing lesions from a previously published dual-energy CEDM study of 120 women to identify enhancing lesion morphologies and to characterize their sizes and margins as detected in CEDM. We have designed a phantom based on these clinical data that incorporates realistic enhancing lesion morphologies for CEDM evaluation. The phantom includes elements of four lesion types observed in CEDM, which broadly follow analogous categories developed from the MRI Breast Imaging, Reporting and Data System (BI-RADS) lexicon. This phantom uses solid iodinated plastic features with accurate iodine concentrations for detection sensitivity experiments. We believe that comparisons of the lesion morphologies through quantitative metrics and reader studies will be useful to test lesion classification and discrimination tasks that can contribute to CEDM performance evaluation.

Melissa L. Hill, Aili K. Bloomquist, Sam Z. Shen, James G. Mainprize, Ann-Katherine Carton, Sylvie Saab-Puong, Serge Muller, Clarisse Dromain, Martin J. Yaffe

Stability of Volumetric Tissue Composition Measured in Serial Screening Mammograms

The purpose of this study is to investigate the categorization and variation of serial mammogram pairs in dense and non-dense classes. When introducing density based stratified screening, the differences in density between screening rounds should be as small as possible to prevent women and clinicians losing confidence in the stratification scheme. A total of 8843 mammogram pairs (current and prior, mean screening interval 22.65 months) were categorized in dense and non-dense cases based on percent density and volume of glandular tissue. The reproducibility of the categories (prior to current) was tested with simple kappa statistics and the causes for a category change were investigated.

When comparing two examinations, the majority of pairs remained in the same category, with

κ

= 0.783 and

κ

= 0.696 based on percent density and glandular tissue volume respectively. For most women, glandular tissue volume and percent density decreases with age. However in 3.2% (4.6%) of the pairs an examination was classified as non-dense followed by dense based on percent density (glandular tissue volume). Natural circumstances can lead to a change in category, for example glandular tissue volume decreases with age, or increases with the use of HRT. However a higher reproducibility in categorization in dense and not-dense classes based on automatic breast density calculations was found, than reported in the literature based on visual assessment. The reproducibility was higher when using percent density for classification.

Katharina Holland, Michiel Kallenberg, Ritse Mann, Carla van Gils, Nico Karssemeijer

Breast Density Classification Based on Volumetric Glandularity Measured by Spectral Mammography

Observations of significant variability in radiologists’ classification of breast density signals the need for objective classification methods. In this study, we develop a model for a radiologist’s BI-RADS classification based on the volumetric glandularity image measured by spectral mammography and a reader study where ten MQSA certified radiologists assigned BI-RADS scores to 300 screening cases. Several combinations of features such as area glandularity based on a certain volumetric glandularity threshold, breast thickness and the spread of glandular tissue were tested as linear classifier parameters. Logistic regression was used to optimize the parameters and cross-validation to assess the agreement with the radiologists’ majority vote, regarded as truth. We show a clear indication that the automatic classification algorithm performs on par with or better than the average individual radiologist.

Henrik Johansson, Miriam von Tiedemann, Björn Cederström

Is Volumetric Breast Density Related to Body Mass Index, Body Fat Mass, Waist-Hip Ratio, Age and Ethnicity for Malaysian Women?

Previous studies have shown that breast cancer has been linked to breast density as well as obesity. We aim to investigate the relationships between body mass index (BMI), body fat mass (BFM), waist-hip ratio (WHR), age and ethnicity with volumetric breast density (VBD) among Malaysian women. In this context, VBD is defined as the ratio of fibroglandular tissue volume to total breast volume. We collected anthropomorphic and body composition data for 2457 subjects undergoing mammographic examination at the University of Malaya Medical Centre, Kuala Lumpur. The data included weight, height, BMI and BFM which were measured with a body composition analyzer. We also measured waist and hip circumferences for 500 of these subjects. A VBD assessment system (Volpara) was used to analyze mammograms. Our results showed that VBD is not significantly correlated with BMI (r

2

= 0.17), BFM (r

2

= 0.19), WHR (r

2

= 0.11). We also noted that VBD is highest among Malaysian women below 40 years old. VBD is highest for Chinese (mean = 11.3%), followed by Malay (mean = 10.1%) and Indian (mean = 9.4%). In conclusion, VBD is dependent on age and ethnicity (ANOVA, p<0.05) but not on BMI, BFM and WHR.

Norhasnah Zakariyah, Kwan-Hoong Ng, Susie Lau, Kartini Rahmat, Farhana Fadzli, Nur Aishah Mohd Taib

Automated Volumetric Breast Density Derived by Statistical Model Approach

Interest is growing in the developing automated breast density measures because of its strong association with breast cancer risk. Although a number of automated methods to quantify mammographic and volumetric density appeared, they still have issues with accuracy and reproducibility; there is demand for developing new accurate and automated breast density estimation techniques. The purpose of this paper is to design and to test a new approach for automatically quantifying true volumetric fibroglandular tissue volumes from clinical screening full-field digital mammograms.

The approach consists in building a statistical model using a training set of digital mammograms with known measures of percent fibroglandular tissue volume, breast volume and fibroglandular tissue volume calculated by phantom based calibration method. To derive these measures, we follow the standard procedure in machine learning: feature generation, feature selection, regression classification of outputs, final model building and testing.

The correlation of features to known volumetric breast volumes was analyzed. In addition, the performance of models created from different groups of features were studied. By building a statistical model with 28 degrees of freedom, we achieved an R

2

=0.83 between the predicted and measured volumetric breast densities for the testing set of 2000 mammograms which were independent of the training set of 2000 images.

Serghei Malkov, Amir Pasha Mahmoudzadeh, Karla Kerlikowske, John Shepherd

Volumetric Breast Density and Radiographic Parameters

The detection of breast cancer relies on high-quality images from digital mammography. Optimal levels of compression force are unknown, and UK national guidelines recommend forces of less than 200N. However, large variations in compression forces exist and may be influenced by the mammography practitioner and the breast size and pain threshold of the patient. This study examined the relationship between breast density and compression force. Women attending for routine breast screening and who had a mammogram taken by the same practitioner on the same equipment were included in the study (n=211). Volumetric density measurements were obtained using Volpara

TM

and details on imaging parameters were obtained from the DICOM headers. There was a strong, positive correlation between compression force and fibroglandular tissue. There was also evidence of a significant positive association between compression force and breast volume which was independent of the volume of fibroglandular tissue present.

Jennifer Khan-Perez, Elaine Harkness, Clare Mercer, Megan Bydder, Jamie Sergeant, Julie Morris, Anthony Maxwell, Catherine Rylance, Susan M. Astley

The Relationship of Volumetric Breast Density to Socio-Economic Status in a Screening Population

Breast cancer incidence has previously been shown to be greater in women of higher socio-economic status (SES), although the picture is complex due to variations in breast cancer risk factors. We have investigated the relationship between one of the strongest risk factors, breast density, with SES in a population of 6398 post- and peri-menopausal women. Volumetric breast density was measured using Quantra

TM

and Volpara

TM

, and SES was based on the Index of Multiple Deprivation (IMD) associated with each woman’s postcode. The mean IMD score was 26.39 (SD 16.7). Our results show a weak but significant association between SES and volumetric breast density; women from more deprived areas have slightly less dense breasts. After controlling for age, BMI and HRT use the relationship remained significant for density measured by Volpara

TM

(gradient -0.01, p <0.005) but not Quantra

TM

(gradient -0.007, p=0.07).

Louisa Samuels, Elaine Harkness, Susan M. Astley, Anthony Maxwell, Jamie Sergeant, Julie Morris, Mary Wilson, Paula Stavrinos, D. Gareth Evans, Tony Howell, Megan Bydder

Use of Volumetric Breast Density Measures for the Prediction of Weight and Body Mass Index

Body Mass Index (BMI) is an important confounding factor for breast density assessment, particularly where a relative measure (percentage density) is used. Since height and weight are not routinely collected at screening, we investigated the relationship between breast and fat volumes computed by Quantra

TM

and Volpara

TM

and weight/BMI in 6898 women for whom self-reported values are available. A significant positive correlation was found between breast volume and fat volume with both weight and BMI. BMI and Volpara

TM

average fat volume showed the strongest positive relationship (r = 0.728, p<0.001). Using these results we predicted weight and BMI for a separate group of women; these showed moderate intraclass correlation (ICC) agreement with self-reported weight and BMI. The strongest relationship was with weight predicted using Quantra

TM

average fat volume (ICC = 0.634, CI = 0.573-0.689, p<0.001), however our results suggest that it is not possible to accurately predict individuals’ weight and BMI from volumetric breast density measures.

Elizabeth O. Donovan, Jamie Sergeant, Elaine Harkness, Julie Morris, Mary Wilson, Yit Lim, Paula Stavrinos, Anthony Howell, D. Gareth Evans, Caroline Boggis, Susan M. Astley

Mammographic Density and Breast Cancer Characteristics

The aim of this research is to investigate, in a screening population, the relationship between mammographic density and tumour characteristics including size, invasiveness and mammographic features. Mammograms of 105 women with screen detected breast cancer were analysed; 111 lesions were identified. Volumetric density measurements were obtained using Quanta

TM

and Volpara

TM

. Histological information was extracted from the screening database and radiological features were assessed by two expert breast radiologists. Statistical analysis was performed using Mann-Whitney U test and Spearman’s rank order correlation. The median percentage density by Volpara

TM

of women with invasive cancers was significantly higher than those with DCIS (6.5

vs

5.0, p =0.046). Similar results were replicated in the Quantra

TM

measurements, however the results were not statistically significant (17

vs

16, p = 0.19). Further analysis showed a significant positive association between whole tumour size and volumetric density for invasive lesions. Architectural distortion was the only mammographic feature associated with a significant difference in percentage density.

Kathy Ren, Elaine Harkness, Caroline Boggis, Soujanya Gadde, Mary Wilson, Yit Lim, Jamie Sergeant, Sigrid Whiteside, Julie Morris, Susan M. Astley

Managing Tiled Images in Breast Density Measurements

Tiled images are sometimes obtained for women with large breasts, which is a limitation of receptor size. In this retrospective HIPAA compliant study, automated breast density measurements for tiled images are compared with full MLO and CC views. Women with tiled views between July and December 2007 followed by full views within 15 months were included. Volumetric breast density (VBD) for tiled MLO views had very good correlation with full views (r = 0.88), while correlation between tiled and full CC views was poor (r = 0.31). VBD for all women requiring tiled CC views was low (<10%). In conclusion, VBD measured from a tiled MLO view is a reasonable substitute for a full MLO measure. Attributable risk of breast density for women requiring tiled CC views may be sufficiently low compared other factors such as high body mass index.

Jennifer Harvey, Olivier Alonzo, Gordon Mawdsley, Taghreed Alshafeiy, Ralph Highnam, Martin Yaffe

Reliability of Breast Density Estimation in Follow-Up Mammograms: Repeatability and Reproducibility of a Fully Automated Areal Percent Density Method

The aim of this study is to evaluate the reliability of the mammographic density estimations in follow-up examinations as measured by using a fully automated density estimation tool in terms of reproducibility and temporal stability. In our previous study, we have developed the fully automated mammographic density estimation method named as SIGMAM, which is based on the prior statistics of mammograms integrated into a novel level set scheme driven by a population-based tissue probability map (PTPM). This scheme was designed to capture the implicit knowledge of experts’ visual systems in which the learned knowledge was modeled as the PTPM, which was shown to provide relatively high correlation coefficient of 0.93 with experts’ estimations in a single equipment study (Senographe 2000D, GE). In this study, we evaluate the reliability of our SIGMAM method in follow-up mammogram examinations with respect to temporal stability and reproducibility. For evaluation of temporal stability, we selected 170 pairs of CC-view mammograms of 170 female patients taken with the same equipment (Senographe 2000D, GE) within one year from the breast cancer screening database in our institute. On the other hand, we collected pairs of mammograms taken with switched equipment: switched from GE (Senographe DS or Essential) to Hologic (Selenia). In total, 53 pairs of CC-view mammograms from 38 female patients taken within one or two months regarding the menstrual cycle were established as a dataset for reproducibility validation. The correlation coefficient of density estimates in temporal stability mammograms was 0.92, while that of the reproducibility mammograms was 0.87. In conclusion, our SIGMAM method showed relatively high reliability in both reproducibility and temporal stability.

Youngwoo Kim, Jong Hyo Kim

Usefulness of a Combination DBT (Digital Breast Tomosynthesis) and Automated Volume Analysis of Dynamic Contrast-Enhanced Breast (DCEB) MRI in Evaluation of Response to Neoadjuvant Chemotherapy (NAC)

We evaluated the usefulness of DBT and automated volume analysis with DCEB MRI to assess its potential role in estimating viable tumor volume in pre-and pos-t NAC images in response to treatment in comparison with FFDM and US. Twenty women having 21 lesions, in total were recruited for this study.The diagnostic procedures were performed within one month prior to surgery. FFDM, DBT, US and DCEB MRI were performed on each of the patients before and after NAC. The imaging data was analyzed by a medical workstation dedicated to breast MRI imaging. Utilizing the dynamic contrast images from 1st to 4th phase, volume statistics with VOI (volume of interest) and the volume was automatically calculated and evaluated as to the efficacy of NAC. DBT has the advantage of providing macroscopic pathological fidings in total without utitiling contrast medium. On the other hand, DCEB MRI has the advantage of providing numerical and detailed vascularity details of viable areas. In accordance with the results, a combination of DBT and automated volume analysis of DCEB MRI will contribute to more accurate diagnosis in the assessment of pathological response to NAC.

Nachiko Uchiyama, Takayuki Kinoshita, Takashi Hojo, Sota Asaga, Minoru Machida, Hitomi Tani, Mari Kikuchi, Yasuaki Arai, Kyoichi Otsuka

Clinical Efficacy of Novel Image Processing Techniques in the Framework of Filtered Back Projection (FBP) with Digital Breast Tomosynthesis (DBT)

Digital breast tomosynthesis (DBT) slices are reconstructed from projections acquired within a limited angular range. Out-of-plane artifacts are inevitable in reconstructed DBT images. In this study, we evaluated novel image processing techniques in the framework of filtered backprojection (FBP) and compared the results with reconstruction using a previously used FBP method. The novel FBP reconstruction has an adapted filter kernel, uses unbinned projections, performs an adaptive collapsing scheme and statistical artifact reduction, and applies iterative filtering in the image domain. Fifty-four image pairs were evaluated by three experienced radiologists. The images were compared on a 7-point scale (-3, -2, -1, 0, +1, +2, and +3) according to the following five categories: (1) visibility of noise, (2) diagnostic certainty regarding masses, (3) diagnostic certainty regarding microcalcifications, (4) visibility of structures in the pectoral muscle, and (5) overall image quality. The results showed a statistically significant superiority of the novel FBP reconstruction in comparison with standard FBP (p < 0.05). In particular, the improvement of the diagnostic certainty related to microcalcifications with the novel FBP is noteworthy.

Nachiko Uchiyama, Minoru Machida, Hitomi Tani, Mari Kikuchi, Yasuaki Arai, Kyoichi Otsuka, Andreas Fieselmann, Anna Jerebko, Thomas Mertelmeier

A Revisit on Correlation between Tabár and Birads Based Risk Assessment Schemes with Full Field Digital Mammography

Mammographic risk assessment is used to determine the probability of a woman developing breast cancer and it plays an important role in the early detection and disease prevention within screening mammography. Tabár and Birads are two fundamentally different risk schemes, one is assessed based on mixtures of breast parenchyma and the other one is assessed based on the percentage of dense breast tissue. This paper presents findings on the correlation between these two mammographic risk assessment schemes; aspects with respect to reader experience and related inter reader variability were also investigated. As a follow up (revisit) investigation to a previously published paper, the new results have shown a strong correlation between Tabár and Birads with the highest Spearman’s correlation coefficient > 0.92 and

κ

 = 0.86% (almost perfect agreement). The statistical results vary with readers’ mammographic reading experience, which also indicated subtle information such as that some mixture of breast parenchma (Tabár specific mammographic building blocks) may be more likely to cause inter reader variability.

Wenda He, Minnie Kibiro, Arne Juette, Erika R. E. Denton, Reyer Zwiggelaar

Predicting Triple-Negative Breast Cancer and Axillary Lymph Node Metastasis Using Diagnostic MRI

Early classification of breast cancers by molecular subtype allows for expeditious characterization of the disease and selection of appropriate treatment options. This ability is especially a concern for “triple-negative” cancers, which lack expression of the three cell surface receptors that most breast cancer hormonal therapies target, tend to be the most aggressive/metastatic compared to other subtypes, have lymph node involvement at diagnoses, and have relatively poor prognoses. In this study, we aim to develop predictive models using Dynamic Contrast-Enhanced (DCE) MRI-extracted features to identify triple-negative cancers and axillary lymph node metastasis at the time of diagnostic imaging. Using only morphological, pharmacokinetic, densitometric, statistical, textural, and textural kinetic features obtained from DCE-MRI, we were able to classify 91.3% of 69 lesions correctly for triple-negative status with a sensitivity of 55.6%, a specificity of 96.7, and an AUC of 0.889; 71.6% of lesions correctly for lymph node metastasis with a sensitivity of 50.0%, a specificity of 82.2%, and an AUC of 0.677.

Jeff Wang, Fumi Kato, Kohsuke Kudo, Hiroko Yamashita, Hiroki Shirato

Understanding the Role of Correct Lesion Assessment in Radiologists’ Reporting of Breast Cancer

Despite the innovations in breast imaging technology, the miss rates of breast cancers at mammography screening have remained stable, ranging from 10-30% per year. While many factors have been linked to radiologist performance (such as volume of cases read, years of experience reading mammograms), little is known about the relationship between the cancers correctly reported by the radiologists and the characteristics of the background and the malignant lesion. In this study we have used the BREAST platform to allow 92 radiologists to read a case set of 60 digital mammograms, of which 20 depicted cancer. Readers were divided in 4 groups, obtained from the quartiles of the median localization sensitivity performance. Median location sensitivity for all readers was 0.71 (IQR=0.21). Statistically significant differences were observed among the groups in correctly reporting several types of lesion; for example, stellate masses were correctly reported only 37.5% by the poorest performers (median location sensitivity < 0.5), vs 88.9% by the top performers (median location sensitivity ≥ 0.92, z=-3.317, P=0.0017). When compared to top performers, the poorest performers had more difficulty reporting smaller lesions (<10mm) (40.9% vs 90.9% from top performers, z=-3.354, P=0.0008). Results suggest a link between the types of lesions more often missed by radiologists and their median location sensitivity.

Claudia Mello-Thoms, Phuong Dung Trieu, Mohammed A. Rawashdeh, Kriscia Tapia, Warwick B. Lee, Patrick C. Brennan

Realistic Simulation of Breast Tissue Microstructure in Software Anthropomorphic Phantoms

Software anthropomorphic breast phantoms have been used in virtual clinical trials for preclinical validation of breast imaging systems. Virtual trial quality depends largely on the realism of the simulated breast anatomy. Our phantom design has been focused on the simulation of large-scale and meso-scale anatomical structures, including the breast outline, skin, and matrix of Cooper’s ligaments and tissue compartments. Realism of such a design has been confirmed in comparative studies of phantom and clinical power spectra and parenchymal texture. We present a novel method for simulating the hierarchical organization of breast tissue subcompartments, seen in detailed histological images. The subcompartmentalization introduces microstructure in breast phantoms, resulting in improved realism of phantom images. The qualitative validation of phantoms with simulated microstructure is discussed in this paper; the quantitative validation in ongoing.

Predrag R. Bakic, David D. Pokrajac, Raffaele De Caro, Andrew D. A. Maidment

A Virtual Human Breast Phantom Using Surface Meshes and Geometric Internal Structures

A highly realistic virtual imaging chain including a model of the human breast could become an important tool for breast imaging system development, optimization and performance assessment. Here we propose a virtual modular breast model with mathematically defined complex anatomical structures that are each represented by a surface mesh. The anatomical structures are designed based on previously published descriptions of internal breast anatomy. Several phantom iterations were performed to tune simulated breast phantom x-ray images visually to real patient images. X-ray image simulation was performed using a polygonal projector. Visual assessment of simulated images of our breast phantoms has shown that our phantom can mimic the range of local features seen in mammograms, contrast-enhanced spectral mammography images and breast CT slices. Further understanding of the fibroglandular tissue structure and its spatial distribution are needed to improve our simulations.

Ann-Katherine Carton, Anthony Grisey, Pablo Milioni de Carvalho, Clarisse Dromain, Serge Muller

Characterisation of Screen Detected and Simulated Calcification Clusters in Digital Mammograms

Simulated microcalcifciation clusters have been used in studies performed to investigate the effect of different imaging conditions on cancer detection in breast screening. This work compares the characteristics of the simulated clusters to screen-detected calcification clusters. Using a database of 271 screen-detected cancers it was found that 67 (25%) presented radiographically as calcification clusters. The characteristics of 1215 microcalcifications from all 67 clusters and 304 microcalcifications from 30 simulated clusters were quantitatively analysed. The diameter of simulated calcifications were within the range of 99% of real calcifications. The cluster diameters of the simulated clusters were within the range of 70% of the real clusters. Our simulated calcifications had similar characteristics to real calcifications but were representative of smaller clusters which represent 17% of screen-detected cancers. Consequently, a significant change in detection of our simulated clusters due to change in imaging condition has a predictable impact on cancer detection in screening.

Lucy M. Warren, Louise Dummott, Matthew G. Wallis, Rosalind M. Given-Wilson, Julie Cooke, David R. Dance, Kenneth C. Young

Development of a Micro-Simulation Model for Breast Cancer to Evaluate the Impacts of Personalized Early Detection Strategies

Breast cancer screening with mammography has been shown to reduce breast cancer mortality. However the frequency and the age range for screening eligibility has been controversial. Individual risk based screening regimens have recently been proposed to overcome some of the weaknesses of screening mammography. However, it is not possible to evaluate the full impact of such risk based individualized screening strategies in Canadian context. Therefore a mathematical cancer control model for breast cancer using care paths and cancer control data from the province of BC is being developed to model different early detection strategies. The model will incorporate the incidence, detection, diagnosis, progression, and case fatality of breast cancer in BC as baseline to make projections of the population health and economic impacts of different early detection methods for breast cancer. Once the model is validated, it will be possible to test early detection pathways and strategies, frequencies and durations, as well as any health care costs associated with detection, diagnosis, treatment and on-going care of breast cancer patients.

Rasika Rajapakshe, Cynthia Araujo, Chelsea Vandenberg, Brent Parker, Stephen Smithbower, Chris Baliski, Susan Ellard, Laurel Kovacic, Melanie Reed, Scott Tyldesley, Gillian Fyles, Rebecca Mlikotic

Modelling Vascularity in Breast Cancer and Surrounding Stroma Using Diffusion MRI and Intravoxel Incoherent Motion

Contrast-enhanced magnetic resonance imaging (MRI) has shown variation in the stroma with distance from the tumor and this correlates with histological microvessel density. To date, however, conventional diffusion MRI has demonstrated limited sensitivity to these changes. This study modelled the diffusion signal by intravoxel incoherent motion (IVIM) to obtain parameters related to the vasculature and tissue diffusion. This revealed a small vascular contribution to the signal in tumor and peri-tumoral stroma within 8 mm. Monoexponential fitting performed worse than the IVIM model in tumor and stroma within 8 mm, but was sufficient in more distal stromal regions where lower microvessel density is expected. Modelling diffusion MRI by IVIM provided a measure of vascularity that may complement information from DCE-MRI and yielded additional information about diffusion in the extravascular tissue.

Colleen Bailey, Sarah Vinnicombe, Eleftheria Panagiotaki, Shelley A. Waugh, John H. Hipwell, Daniel C. Alexander, Kathryn Kitching, Patsy Whelehan, Sarah E. Pinder, Andrew Evans, David J. Hawkes

Monte Carlo Modeling of the DQE of a-Se X-Ray Detectors for Breast Imaging

We study the resolution characteristics of a-Se semiconductor x-ray detectors using ARTEMIS, a detailed Monte Carlo transport code that simulates the three-dimensional spatial and temporal transport of electron-hole pairs under an external electric field. The model takes into account generation and re-absorption of characteristic x rays, spreading due to Compton scattering and high-energy secondary electron transport, and drift and diffusion of electron-hole pairs under applied bias. The point responses for a 200

μ

m thick mammography detector for RQA radiation qualities are simulated using parallel processing. Line spread functions and modulation transfer functions show a dependence on incident x-ray energy and spatial frequency.

Yuan Fang, Andreu Badal, Karim S. Karim, Aldo Badano

kVp Tool for Digital Mammography Using Commercial Metallic Foils

An alternative method to evaluate the kVp on digital mammography units was developed using commercial metallic foils of different elements and an aluminum step wedge contained in a mammography test phantom developed at National University of México (UNAM) as a low cost tool. Relative response of metallic foils

(Cu+5Al+8Al) / (Mo+Rh+Ag) vs kVp

, for a numerical analysis and in experimental method of three FFDM systems, shows a linear behavior and permits to calculate kVp with precision of ±0.4 kV. First results are shown and further work is still in process.

Héctor A. Galván, Yolanda Villaseñor

Possibility of Exposure Dose Reduction in Contrast Enhanced Spectral Mammography Using Dual Energy Subtraction Technique : A Phantom Study

Contrast Enhanced Spectral Mammography using energy subtraction technique (CESM) with iodinated contrast media is promising technology to improve the contrast of the image and detectability of tumor. One of the issues to be solved on this technique is giving additional radiation exposure to the patient compared with conventional mammography. We investigated phantom study to optimize scan protocol and parameter setting for reducing radiation dose without image degradation. We acquired the images with two different imaging conditions, fully-automated mode and manual mode, and evaluated image quality by image noise, contrast and exposure dose. On image quality evaluation in manual mode, the normalized noise power spectrum (NNPS) at low-energy image was increased and image quality became worse, but the quality of recombined image was not significantly different comparing to fully-automated mode. The contrast at low-energy image in manual mode was slightly deteriorated, but at recombined image was not much different comparing to auto mode. On the other hand, average glandular dose (AGD) was able to be reduced to 1.41 mGy from 1.96 mGy by setting manual mode. These results suggest it may possible to reduce the exposure dose by using manual mode instead of fully-automated mode when CESM has performed in clinical service.

Noriko Nishikawa, Kaori Yanagisawa, Kuniji Naoi, Yutaka Ohnuma, Yoshihisa Muramatsu

A Protocol for Quality Control Testing for Contrast-Enhanced Dual Energy Mammography Systems

Physics and radiographer QC procedures are urgently needed as the first few contrast-enhanced dual energy systems have been installed in the U.K. Preliminary work on one commercially available system has enabled us to propose new tests, relevant to the properties of dual energy imaging systems. Results are presented for measurements with the chosen phantom, which contains disks with a range of iodine content from 0.25 to 2 mg/cm

2

. Breasts of different thicknesses and different glandularity were simulated by adding slabs of CIRS material, of a range of compositions, on top of the phantom. The system tested had a response which was proportional to the iodine content of disks in the phantom, had good reproducibility, and did not change significantly when simulated breast thickness and composition were varied.

Jennifer Oduko, Peter Homolka, Vivienne Jones, David Whitwam

Trends in Mammogram Image Quality, Dose and Screen-Detected Cancer Rates in an Organized Screening Mammography Program

BACKGROUND: The Screening Mammography Program of British Columbia (SMPBC), Canada is a population based program that regularly performs quality assurance testing and outcomes analysis.

METHODS: A study was conducted to analyze the trends in the SMPBC quality assurance data from 1994 onwards to investigate any correlation between improvements in image quality (IQ), changes in radiation dose delivered per screen and detection of breast cancers.

RESULTS: Both IQ and cancer detection rates of invasive tumours ≤5 mm increased over 1994-2011 which came at the cost of a 127% increase in radiation dose delivered to the breast between 1994-2005 (IQ increased 21%, tumours ≤5 mm increased 107%). In subsequent years, as digital units started to replace film units the programs’ average IQ and CDRs remained unchanged, while the integration of digital units reduced the dose delivered at a populational level.

CONCLUSION: Improvements in IQ coincided with increased detection of small tumours.

Brent Parker, Rasika Rajapakshe, Ashley Yip, Teresa Wight, Nancy Aldoff, Janette Sam, Christine Wilson

Power Spectrum Analysis of an Anthropomorphic Breast Phantom Compared to Patient Data in 2D Digital Mammography and Breast Tomosynthesis

Digital breast tomosynthesis (DBT) images of a novel anthropomorphic breast phantom (UPenn phantom) acquired on two breast tomosynthesis systems were analyzed in terms of their power spectra (PS). The

β

and

κ

power law coefficients were estimated from 2D planar, tomosynthesis projection images and reconstructed planes. These data were compared to the PS characteristics as retrieved from a group of patient data. Power spectra of the UPenn phantom images were very similar to the patient data, with power law parameters in the range of values found in patients. Power law exponents were 2.99 and 3.45 for 2D, 2.87 and 2.75 for DBT projections and, 1.92 and 3.10 for DBT reconstructions for the Siemens and Hologic system respectively. The agreement was better than with other (non-anthropomorphic) 3D structured phantoms, making this phantom a good candidate test object for DBT performance testing.

Lesley Cockmartin, Predrag R. Bakic, Hilde Bosmans, Andrew D. A. Maidment, Hunter Gall, Moustafa Zerhouni, Nicholas W. Marshall

Contrast-Enhanced Digital Mammography Image Quality Evaluation in the Clinic

Some limitations of mammography that particularly affect diagnosis of women with dense breasts, such as tissue superposition and marginal cancer image contrast, can be overcome with the use of contrast-enhanced digital mammography (CEDM). CEDM uses iodinated contrast agents to increase attenuation in areas exhibiting hyper-vascularization, potentially due to tumour angiogenesis, and image subtraction to cancel normal tissue signal. Here, we propose a method for objective task-based image quality evaluation of CEDM that can be routinely carried out in the clinic. A phantom was designed with features that allow for practical measurements of MTF, NPS, and iodine contrast that were used to estimate a CEDM detectability index for a given imaging task. We present results from several months of weekly testing of a commercial dual-energy CEDM system. From these data, we demonstrate measurement sensitivity to variations from standard acquisition conditions, suggesting the potential to identify system failure modes using this approach.

Melissa L. Hill, Aili K. Bloomquist, Sam Z. Shen, James G. Mainprize, Ann-Katherine Carton, Sylvie Saab-Puong, Serge Muller, Martin J. Yaffe

BREAST: A Novel Strategy to Improve the Detection of Breast Cancer

Early diagnosis of breast cancer is highly dependent on quality breast imaging and precise image interpretation. The BREAST programme is an innovative strategy for reader performance self-evaluation in breast cancer detection. Using an online system, detailed feedback on reader/image interpretation is given instantly. Our strategy is currently focused on mammograms but has the potential to be available for a wide range of medical imaging modalities. BREAST also serves a solution to researchers requiring large observer numbers by facilitating the involvement of experts wherever they are located. In summary, BREAST improves the efficacy of mammographic cancer detection through a system of reader performance monitoring and enables research studies with a large amount of robust data.

Patrick C. Brennan, Phuong Dung Trieu, Kriscia Tapia, John Ryan, Claudia Mello-Thoms, Warwick Lee

A Regional Web-Based Automated Quality Control Platform

Quality control is a key factor in ensuring a high standard of care in the field of mammography. We have found that abrupt irregularities in image quality from mammography units can arise as the result of factors ranging from vendor software upgrades, having software parameters modified during unit maintenance, or even having detectors replaced. We have developed both a simple weekly quality control test performed on processed images that can quickly capture these changes in image quality, as well as a centralized software platform that automates our test across several mammography centers. Technologists acquire a phantom exposure and upload it to our regional PACS network. The images are then automatically downloaded, analysed, and the results stored by the mammoQC software. These results are instantly available to technologists via a web dashboard, where reports can be generated automatically. Our platform currently services over 25 locations in British Columbia.

Stephen Smithbower, Rasika Rajapakshe, Janette Sam, Nancy Aldoff, Teresa Wight

A European Protocol for Technical Quality Control of Breast Tomosynthesis Systems

Quality control (QC) procedures for digital breast tomosynthesis (DBT) systems are a crucial part of the acceptance of a new modality. In contrast to the situation in the US, the European approach is to provide a device independent protocol with limiting values that should be applied to all systems. A European QC protocol that deals with this challenge is being developed and is currently work-in-progress. In this paper four specific QC tests for DBT, which have reached an (almost) final stage, are presented: reproducibility, system projection MTF, z-resolution and missed tissue at the top and bottom of the reconstructed volume. The proposed tests have been evaluated on several DBT systems. The encouraging results show that these tests will form an appropriate and necessary part of QC procedures for DBT.

Ruben E. van Engen, Hilde Bosmans, Ramona W. Bouwman, David R. Dance, Patrice Heid, Barbara Lazzari, Nicholas W. Marshall, Stephan Schopphoven, Celia Strudley, Martin Thijssen, Kenneth C. Young

Conventional Mammographic Image Generation Method with Increased Calcification Sensitivity Based on Dual-Energy

The visualization of calcifications could be obscured in mammograms because of overlapping of tissue structures. Dual-energy digital mammography (DEDM) can generate tissue-subtracted image for improving the detectability of breast calcifications, but the mass information is missing. This paper proposes a conventional mammographic image generation method with increased calcification sensitivity based on DEDM. Firstly, a conventional mammographic image is generated with low-energy and high-energy images based on multi-scale decomposition and reconstruction. Secondly, the tissue-subtracted “calcification image” is generated using a nonlinear inverse mapping function with calcification pixels marked. Finally, the density values of the marked calcification pixels in the reconstructed mammographic image are increased for better visualization. Preliminary results show that the proposed DEDM method can generate both calcification and conventional mammogram-like images and the calcification sensitivity is increased. The CNR of calcifications of 50% glandular ratio has been increased from 2.75 to 9.32.

Xi Chen, Xuanqin Mou

Development of Mammography System Using CdTe Photon Counting Detector for Exposure Dose Reduction

- Study of Effectiveness of the Spectrum by Simulation -

We have proposed a new mammography system using a cadmium telluride photon counting detector to reduce exposure dose. In conjunction with this, we propose a new high x-ray energy spectrum with tungsten/barium (W/Ba) as a target/filter. In this study, the usefulness of the W/Ba spectrum, in terms of image quality and dose distribution is evaluated through Monte Carlo simulation. The contrast-to-noise ratio and dose distribution are measured using polymethyl methacrylate phantoms of 2, 4, and 7 cm thickness. In each case, the result obtained using the W/Ba spectrum is better than that from conventional mammographic spectra. The results of this study indicate that, by using a higher energy x-ray than in conventional mammography, it is possible to obtain significant exposure dose reduction without loss of image quality.

Naoko Niwa, Misaki Yamazaki, Sho Maruyama, Yoshie Kodera

Development of Mammography System Using CdTe Photon-Counting Detector for Exposure Dose Reduction

- Evaluation of Image Quality in the Prototype System -

We discuss a new mammography system using a cadmium telluride (CdTe) photon-counting detector for exposure dose reduction. We created a prototype system that uses a CdTe detector and an automatic moving stage. For a variety of conditions, we measured the image properties and evaluated the image quality produced by reconstructing scanning images from several thousand frame data elements obtained by shift-and-add method, from which it was demonstrated that the basic detector performance in terms of output linearity with respect to X-ray intensity was good. The spatial resolution under various conditions and the linearity of the relationship between the thickness of the acrylic step and reconstructed scanning images were also measured. Finally, we evaluated the image quality obtained by scanning a breast phantom. Our results show that the developed prototype system can improve image quality by optimizing the balance between the shifting-and-adding operation and the output of the X-ray tube.

Misaki Yamazaki, Niwa Naoko, Sho Maruyama, Yoshie Kodera

Investigation of Dependence on the Object Orientation in Visibility-Contrast Imaging with the X-Ray Talbot-Lau Interferometer

Visibility-contrast image obtained by the X-ray grating interferometer reflects reduction of coherence due to the object’s structures. In the case of the one-dimensional grating, visibility-contrast image is affected by relative angle between the structures of the object and the grating. In this study, we have investigated the features of the visibility-contrast signal at the edge of the object. We imaged the acrylic cylinder with the Talbot-Lau interferometer (Konica Minolta, Inc.) by rotating from -90 degrees to +90 degrees with respect to the grating’s periodic direction, and measured its edge signal. The signal became its maximum at -90 degrees and +90 degrees, and became zero at 0 degree. This result showed a good agreement with the angle dependency of the x-ray refraction at the edge of the cylindrical structure.

Takayuki Shibata, Shohei Okubo, Daiki Iwai, Junko Kiyohara, Sumiya Nagatsuka, Yoshie Kodera

Development of New Imaging System Based on Grating Interferometry : Preclinical Study in Breast Imaging

A new imaging system based on an x-ray Talbot-Lau interferometry was developed. The preclinical study with mastectomy specimens was conducted, and the three types of images, i.e., the attenuation contrast(ATT) image, the differential phase contrast(DPC) image, and the x-ray small angle scattering(SAS) image, obtained by the system were compared to the pathological result. As a result, the SAS image showed micro-calcifications clearly. On the other hand, the inside of the mass with invasive carcinoma was visualized with relatively lower signal. The SAS image seemed to correspond to the homogeneity of the breast tissues. The breast images obtained with Talbot-Lau interferometry showed the different aspects which cannot be depicted with the conventional x-ray image. Comparative reading of the three images would enable us to get additional information of breast tissues.

Tokiko Endo, Shu Ichihara, Suzuko Moritani, Mikinao Ooiwa, Misaki Shiraiwa, Takako Morita, Yasuyuki Sato, Junko Kiyohara, Sumiya Nagatsuka

Basic Study on the Development of a High-Resolution Breast CT

X-ray breast computed tomography (breast CT) was developed by some research groups to overcome the limitations of mammography. Breast CT is expected to be an effective diagnostic tool because it can generate three-dimensional images of a breast. However, the spatial resolution of the existing system is not satisfactory for identifying microcalcifications within the breast. The purpose of this study was to develop a prototype of high-resolution breast CT system, and to evaluate the imaging properties of the developed system. Our experimental system consists of a microfocus X-ray tube and a flat panel detector with a C-arm frame, a bed, and their controllers. Images were reconstructed by using cone-beam X-ray projections and the Feldkamp-Davis-Kress algorithm. We used phantoms to experimentally evaluate three imaging properties and exposure dose. Consequently, the modulation transfer function value was 0.1 at the frequency of 6.0 LP/mm, which is higher than that of clinical CT and breast CT. Breast phantom microcalcifications were observed clearly. Furthermore, entrance surface dose in the experimental system was similar to that of mammography. These results indicate that our experimental system overcomes the limitations of both the mammogram and existing breast CT systems.

Atsushi Teramoto, Tomoyuki Ohno, Fumio Hashimoto, Chika Murata, Keiko Takahashi, Ruriha Yoshikawa, Shoichi Suzuki, Hiroshi Fujita

Analysis of Dependence of Detector Position on Detected Scatter Distribution in Dedicated Breast SPECT

In SPECT, scattered photons contribute to the detected signal, reducing contrast and quantification accuracy. Several methods exist to correct scatter, including the dual-energy window technique, but have not been fully evaluated on non-traditional SPECT trajectories. Using MCNP5, a Monte Carlo study was performed to analyze how incident scatter is affected by detector position for breast SPECT. An ideal detector was positioned at various azimuthal and polar angles relative to a pendant breast geometry. Detected scatter from the breast, heart, liver, torso, and lesion was linearly fit; the slope was used to characterize the distribution. Typical photopeak and scatter energy window ratios were calculated. Results indicate detected scatter depends upon detector position and its vantage of major uptake organs; however, the effect is minimal for non-direct views, with a ratio of 0.37. A single coefficient for dual-energy window scatter correction should suffice for breast imaging trajectories ignoring direct views of the heart/liver.

Steve D. Mann, Jainil P. Shah, Martin P. Tornai

Evaluation of Physical and Psychological Burden of Subjects in Mammography

The realities of physical and psychological burden associated with mammography are not fully understood. We have measured the muscle activity and the sympathetic nervous activity of subjects during mammography to estimate the burden. The experimental results suggested that positioning during mammography affects the muscle activity and the sympathetic nervous activity of the body. We carried out another preliminary experiment for decreasing the examinee’s burden using humorous video. In the experiment, two groups ("humor group" and "neutral group") underwent mammography. The humor group was shown a humorous video during mammography. As a result, numerical rating scale scores of humor group on pain and experience time were higher than that of neutral group (no significant difference). In conclusion, the physical and psychological burden of mammography examinees could be evaluated by measuring the muscle activity and the sympathetic nervous activity. Humorous video may be effective at increasing pain tolerance of subjects during mammography.

Yongbum Lee, Mieko Uchiyama

Mammographic Image Database (MIDB) and Associated Web-Enabled Software for Research

Current efforts relating to the uptake, evaluation and research into digital medical imaging require the large-scale collection of images (both unprocessed and processed) and data. This demand has led us to design and implement a flexible mammographic image repository, which prospectively collects images and data from multiple screening sites throughout the UK. The MIDB has been designed and created to provide a centralised, fully annotated dataset for research purposes. One of the most important features is the inclusion of unprocessed images. In addition to the images and data, systems have been created to allow expert radiologists to annotate the images with interesting clinical features and provide descriptors of these features. MedXViewer (Medical eXtensible Viewer) is an application we have designed to allow workstation-independent, PACS-less viewing and interaction with anonymised medical images (e.g. for observer studies). With these integrated tools, the MIDB has become a valuable resource for running remote observer studies and providing data and statistics for imaging based-research projects. Previously, studies were run by laborious transfers of images to PACS at remote sites and paper-based data manually curated into databases. Apart from the inconvenience, these approaches also suffer from a lack of accurate location information from the paper-based forms.

Mark D. Halling-Brown, Pádraig T. Looney, Mishal N. Patel, Lucy M. Warren, Alistair Mackenzie, Kenneth C. Young

Optimizing High Resolution Reconstruction in Digital Breast Tomosynthesis Using Filtered Back Projection

In Digital Breast Tomosynthesis, a 3D representation of the breast is reconstructed from low-dose projection images acquired over a limited angular range. Each such image contains high level of noise which is often counteracted by a projection binning to yield the CNR desired in clinical applications. However, this approach reduces spatial resolution and makes imaging of high frequency structures such as micro-calcifications challenging. In this paper, we describe a Filtered Back Projection (FBP) reconstruction method optimized to yield improved CNR without sacrificing spatial resolution. The results from our quantitative evaluation and clinical reading by experienced radiologists indicate that the proposed methods can significantly improve contrast and sharpness of micro-calcifications and reduce noise compared to a baseline FBP method with standard filter settings.

Shiras Abdurahman, Frank Dennerlein, Anna Jerebko, Andreas Fieselmann, Thomas Mertelmeier

The Investigation of Different Factors to Optimize the Simulation of 3D Mass Models in Breast Tomosynthesis

The development of 3D mass models of different shapes, margins and degrees of malignancy may allow more profound and clinically relevant testing and optimization of the performance of the newly introduced 3D modalities such as breast tomosynthesis and breast-CT. Three dimensional mass models had been developed earlier and were validated for the realism of their appearance after simulation into 2D and tomosynthesis patient images. Based on the feedback of the readers and the results of the simulations of the earlier study we initiated the present study in which we investigated the effect of insertion position and background glandular tissue estimation on the appearance of these masses. A subset of these masses was re-simulated in another position and using a different background estimator. These simulated masses were subsequently evaluated by an experienced radiologist on a 5-point scale realism score. The results showed that the insertion position of simulated masses is a significant factor in the appearance of realism of these masses and careful choices should be made.

Eman Shaheen, Frédéric Bemelmans, Chantal Van Ongeval, Frederik De Keyzer, Nausikaä Geeraert, Hilde Bosmans

Clinical Evaluation of Dual Mode Tomosynthesis

Clinical performance achieved by adding interpretation of Tomosynthesis images to 2D images (hereinafter 2D+Tomo) was studied. 100 cases who gave written informed consent (ST mode: angular range ±7.5deg, 50 cases and HR mode: ±20deg, 50 cases) were obtained and 7 radiologists interpreted all images. In ST, the sensitivity is significantly increased by 15% (P<.01) and specificity is equivalent (P=.73). In HR, sensitivity of 2D+Tomo against 2D alone is significantly increased by 30% (P<.001) and specificity is significantly decreased by 5% (P<.01). ST, which has higher sensitivity and equivalent specificity, can be used for screening, and HR, which can visualize structures such as lesions in details, can be used for diagnosis. Moreover, specificity enhancement due to inhibition of false positive and sensitivity enhancement are confirmed by trial for appropriate segmentation in Japanese category classification C3. Further detailed clinical performance will be studied such as in ROC analysis with more cases.

Tokiko Endo, Mikinao Ooiwa, Takako Morita, Namiko Suda, Kazuaki Yoshikawa, Misaki Shiraiwa, Yukie Hayashi, Takao Horiba, Yasuyuki Sato, Syu Ichihara, Tomonari Sendai, Tetsuro Kusunoki, Takahisa Arai

Image Quality of Thick Average Intensity Pixel Slabs Using Statistical Artifact Reduction in Breast Tomosynthesis

Digital Breast Tomosynthesis (DBT) has the potential to replace or supplement Digital Mammography (DM). Studies have shown that it takes radiologists more time to read DBT examinations compared with DM. The slice separation of image volumes has been set to 1 mm on most systems. By using thicker slices review time could be reduced. This paper investigates the possibility of using 2 mm Average Intensity Pixel (AIP) slabs for image review. The thicker slabs were created using a method based on statistical artifact reduction and super-resolution. Six radiologists were presented with 20 sets of images containing 16 tumor masses and 8 micro-calcification clusters. They ranked 2 mm slabbed sets relative to standard 1 mm. Visibility (P = .0044) of micro-calcifications improved and there was no significant effect on mass visibility (P = .46). The results indicate that it is possible to review DBT-volumes with 2 mm slabs without compromising image quality.

Magnus Dustler, Pontus Timberg, Anders Tingberg, Sophia Zackrisson

Detection of Spiculated Lesions in Digital Mammograms Using a Novel Image Analysis Technique

We have applied novel computational image analysis algorithms to detect malignant masses in mammograms. Our analysis focuses on spiculated lesions, which are particularly challenging for computer-aided detection methods. The algorithm uses the principle of locally-normalised correlation coefficients to identify patterns of motifs representing a spiculated feature. A combination of correlation maps indicating the maximum correlation of the motif at each position relative to the mammogram, and of the pattern of angles for which this maximum is observed, are used to locate spiculated lesions in a verified test dataset. The test set of images has been annotated by an expert reader, and allows objective evaluation of computer-aided detection procedures. In a blind test using an automated procedure our method identified 54% of the lesion locations in the set of test images. This initial blind testing and comparison with expert annotated images, representing a ground truth, indicates feasibility for our approach. Optimisation of the procedure is expected to yield improved performance.

Ashley Seepujak, Tomas Adomavicius, Sergey Dolgobrodov, Emmanouil Moschidis, Xin Chen, Anthony Maxwell, Susan M. Astley, Alan M. Roseman

Spatial Correlation Analysis of Mammograms for Detection of Asymmetric Findings

We present a novel method to detect asymmetry in mammograms based upon bilateral analysis of the spatial distribution of density within paired mammographic strips. Various differential measures of spatial correlation of gray-scale values were computed with reference to the position of the nipple for a set of 128 pairs of mammograms from the Digital Database for Screening Mammography (DDSM). Features were selected by stepwise logistic regression and the leave-one-patient-out method was used for cross-validation of results. An area under the receiver operating characteristic curve of 0.87 (

SE

 = 0.08) was achieved by using an artificial neural network classifier with radial basis functions.

Paola Casti, Arianna Mencattini, Marcello Salmeri, Rangaraj M. Rangayyan

Temporal Breast Cancer Risk Assessment Based on Higher-Order Textons

Higher-order texture features from 100 mammographic images with known cancer were compared to texture features from 100 images from women with no known cancer. Texture features from images of the same breasts from screening rounds two and four years previously were also compared. The

A

z

score for classifying cancer images from non-cancer images was 0.749. The

A

z

score for classification two years previous to detection of cancer was 0.674 and the score for four years previous was 0.601. There was no signicant difference between classifying images from the round in which cancer was actually detected and the screening rounds two and four years previous. Similar results were obtained if the breast with no known cancer (contralateral breast) was used instead the breast with cancer, leading to the conclusion that texture alone has moderate predictive power regarding breast cancer risk and that this predictive value is roughly constant in the four years prior to mammographically apparent cancer.

Xi-Zhao Li, Simon Williams, Peter Downey, Murk J. Bottema

Invariant Features for Discriminating Cysts from Solid Lesions in Mammography

Feature extraction is an integral of all Computer Aided Diagnosis (CAD) systems. Due to the presence of fibroglandular tissue however, measurements are perturbed by unwanted influences and therefore, the same descriptor will yield different values for different amounts of occluding structures. To aid the statistical learning used for classification, we need to design features that are

invariant

to unwanted influences. In this paper, we propose a simple model of the tumour and its surrounding tissue and show how this model can be used to derive descriptors that are invariant to obscuring tissue, rather than heuristically defining a set of descriptors, which is common practice in many CAD papers. We tailor the descriptors to optimally discriminate between tumours and cysts, by assuming a parametric form of the lesions. Results show a significant discriminative improvement over simple, more commonly used contrast features and we obtained an AUC of 0.77 using both CC and MLO images.

Thijs Kooi, Nico Karssemeijer

Breast Masses Identification through Pixel-Based Texture Classification

Mammographic image analysis plays an important role in computer-aided breast cancer diagnosis. To improve the existing knowledge, this paper proposes a new efficient pixel-based methodology for tumor vs non-tumor classification. The proposed method firstly computes a Gabor feature pool from the mammogram. This feature set is calculated through multi-sized evaluation windows applied to the probabilistic distribution moments, in order to improve the accuracy of the whole system. To deal with a high dimensional data space and a large amount of features, we apply both a linear and non-linear pixel classification stage by using Support Vector Machines (SVMs). The randomness is encoded when training each SVM using randomly sample sets and, in consequence, randomly selected features from the whole feature bank obtained in the first stage. The proposed method has been validated using real mammographic images from well-known databases and its effectiveness is demonstrated in the experimental section.

Jordina Torrents-Barrena, Domenec Puig, Maria Ferre, Jaime Melendez, Lorena Diez-Presa, Meritxell Arenas, Joan Marti

Automated Labeling of Screening Mammograms with Arterial Calcifications

For the automatic detection of malignant microcalcification clusters in screening mammograms a computer aided detection (CADe) system has been developed. The most frequent false positives of this system are breast arterial calcifications (BACs). The purpose of this study was to construct a method for selecting cases with BACs in mammographic screening data as part of a procedure to reduce false positives of the CADe system. To automatically select cases containing BACs, a GentleBoost classifier was trained. For composing the training set, the CADe system was applied on 10,000 normal cases. From these cases, 400 cases with the most significant false positives were included in the training set and an additional 200 cases with less obvious false positives. For testing, an independent test set was created by cluster detection of 1,000 normal cases and 95 malignant cases. After cluster detection 342 normal cases contained false positives and in 93 malignant cases true positive clusters were detected. In the training set, 244 cases showed signs of BACs and in the test set 95 cases. A total of 102 case-based features were calculated to train the classifier. A ROC curve was calculated of the classification of the test set bootstrapped 5000 times. The area under the curve of the ROC was 0.92 and already 44% of the cases with BACs were detected without any false positives. Furthermore, 90% of the cases with BACs were detected at a false positive rate of 20%. The performance of the proposed selection method implies a good feasibility to classify cases with BACs at high specificity. By using this selection we will be able to apply dedicated methods for false positive reduction due to BACs.

Jan-Jurre Mordang, Jakob Hauth, Gerard J. den Heeten, Nico Karssemeijer

False Positive Reduction in CADe Using Diffusing Scale Space

Segmentation is typically the first step in computer-aided-detection (CADe). The second step is false positive reduction which usually involves computing a large number of features with thresholds set by training over excessive data set. The number of false positives can, in principle, be reduced by extensive noise removal and other forms of image enhancement prior to segmentation. However, this can drastically affect the true positive results and their boundaries. We present a post-segmentation method to reduce the number of false positives by using a diffusion scale space. The method is illustrated using Integral Invariant scale space, though this is not a requirement. It is quite general, does not require any prior information, is fast and easy to compute, and gives very encouraging results. Experiments are performed both on intensity mammograms as well as on Volpara® density maps.

Faraz Janan, Sir Michael Brady, Ralph Highnam

Automated Detection of Architectural Distortion Using Improved Adaptive Gabor Filter

Architectural distortion in mammography is the most missing finding for radiologists, despite high malignancy. Many research groups have developed methods for automated detection of architectural distortion. However, improvement of their detection performance is desired. In this study, we developed a novel method for automated detection of architectural distortion in mammograms. To detect the mammary gland structure, we used an adaptive Gabor filter, whichconsists of three Gabor filters created by changing the combination of parameters. The filter that is best matched to the mammary gland structure pixel by pixel in the mammogram is selected. After detecting the mammary gland, enhancement of the concentrated region and false positive reduction are performed. In the experiments, we verified the detection performance of our method using 50 mammograms. The true positive rate was found to be 82.45%, and the number of false positive per image was 1.06. These results are similar to or better than those of existing methods. Therefore, the proposed method may be useful for detecting architectural distortion in mammograms.

Ruriha Yoshikawa, Atsushi Teramoto, Tomoko Matsubara, Hiroshi Fujita

Detecting Abnormal Mammographic Cases in Temporal Studies Using Image Registration Features

Image registration is increasingly being used to help radiologists when comparing temporal mammograms for lesion detection and classification. This paper evaluates the use of image and deformation features extracted from image registration results in order to detect abnormal cases with masses. Using a dataset of 264 mammographic images from 66 patients (33 normals and 33 with masses) results show that the use of a non-rigid registration method clearly improves detection results compared to no registration (AUC: 0.76 compared to 0.69). Moreover, feature combination using left and right breasts further improves the performance (AUC to 0.88) compared to single image features.

Robert Martí, Yago Díez, Arnau Oliver, Meritxell Tortajada, Reyer Zwiggelaar, Xavier Lladó

Analysis of Mammographic Microcalcification Clusters Using Topological Features

In mammographic images, the presence of microcalcification clusters is a primary indicator of breast cancer. However, not all microcalcification clusters are malignant and it is difficult and time consuming for radiologists to discriminate between malignant and benign microcalcification clusters. In this paper, a novel method for classifying microcalcification clusters in mammograms is presented. The topology/connectivity of microcalcification clusters is analysed by representing their topological structure over a range of scales in graphical form. Graph theoretical features are extracted from microcalcification graphs to constitute the topological feature space of microcalcification clusters. This idea is distinct from existing approaches that tend to concentrate on the morphology of individual microcalcifications and/or global (statistical) cluster features. The validity of the proposed method is evaluated using two well-known digitised datasets (MIAS and DDSM) and a full-field digital dataset. High classification accuracies (up to 96%) and good ROC results (area under the ROC curve up to 0.96) are achieved. In addition, a full comparison with state-of-the-art methods is provided.

Zhili Chen, Harry Strange, Erika Denton, Reyer Zwiggelaar

Differentiation of Malignant and Benign Masses on Mammograms Using Radial Local Ternary Pattern

Texture information of breast masses may be useful in differentiating malignant from benign masses on digital mammograms. Our previous mass classification scheme relied on shape and margin features based on manual contours of masses. In this study, we investigated the texture features that were determined in regions automatically selected from square regions of interest (ROIs) including masses. As a preliminary investigation, 149 ROIs including 91 malignant and 58 benign masses were used for evaluation by a leave-one-out cross validation. The local ternary pattern and local variance were determined in sub regions with the high contrast and a core region. Using an artificial neural network, the classification performance of 0.848 in terms of the area under the receiver operating characteristic curve was obtained.

Chisako Muramatsu, Min Zhang, Takeshi Hara, Tokiko Endo, Hiroshi Fujita

Statistical Temporal Changes for Breast Cancer Detection: A Preliminary Study

In this paper, we propose a statistical temporal change scheme for early breast cancer detection. Temporal mammographic data have been found useful to detect changes in the breasts of women. Many temporal analysis approaches require temporal image registration. Variations in patient positioning, changes in the field of view and natural changes in the breasts over time pose significant challenges. Our proposed scheme, on the other hand, does not depend on image registration. Instead, the temporal statistical region merging technqiue was used to find homogeneous breast regions over time. Changes identified are then assessed for abnormality by a rule-based classifier. Using a small temporal dataset of 10 women (5 cancerous and 5 normal), the detection rate was found to be 100% with a 0.1 false positive per case (that is, only one false positive was found in the entire dataset of 10 cases). These preliminary results show that the proposed temporal changes detection scheme has a great potential in providing clinical assistance in early breast cancer detection. The results, however, need to be further verified with a larger dataset.

Gobert N. Lee, Mariusz Bajger

Comparison of Calcification Cluster Detection by CAD and Human Observers at Different Image Quality Levels

Previous studies have compared the performance of human observers to the performance of human observers using CAD. Here we compare the performance of human observers to Hologic’s ImageChecker CAD system using a set of 162 images with simulated calcification clusters. The quality of the images was reduced to create four other image sets at different image qualities. These were analysed by the CAD system and the relevant information from the resulting DICOM structured reports was parsed. At the highest image quality level the figure of merit for the CAD was 0.82 and 0.84 for the humans. At the lowest image quality level the figure of merit for the CAD and humans were 0.62 and 0.55 respectively. At each image quality level there was no significant difference (p>0.05). The effect of changes in image quality on calcification detection was similar for human observers and the CAD system.

Padraig T. Looney, Lucy M. Warren, Susan M. Astley, Kenneth C. Young

A Novel Image Enhancement Methodology for Full Field Digital Mammography

During breast screening it is necessary and essential to compress the breast with a compression paddle, in order to obtain a clear mammographic image. The quality of the image has a direct correlation with the accuracy of mammogram reading, which in turn could affect radiologist’s interpretation. Clinical observation has indicated that breast compression may have a side effect on image quality during the image acquisition and can result in unexpected variations in texture and intensity appearances, between breast tissue near the skinline and the rest of the breast. Within computer aided mammography, such variations increase the difficulty in breast tissue modelling and can be detrimental to image analysis, leading to incorrect prompts which can have an impact on sensitivity and specificity of screening mammography. We present an automatic image enhancement approach, in which both Cranio Caudal and Medio-Lateral Oblique views are utilised. We estimate the relative breast thickness ratio at a given projection location in order to alter/correct an inconsistent intensity distribution as a means of improving mammographic image quality. Our dataset consists of 360 full field digital mammographic images was used in a quantitative and qualitative evaluation. Visual assessment indicated good and consistent intensity variation over the processed images, whilst texture information (breast parenchymal patterns) was preserved and/or enhanced. By improving the consistency of the intensity distribution on the mammographic images, the developed method has demonstrated a potential benefit in density based mammographic segmentation and risk assessment. This in turn can be found useful in computer aided mammography, and is beneficial in a clinical setting by aiding screening radiologists in the process of decision making.

Wenda He, Minnie Kibiro, Arne Juette, Erika R. E. Denton, Peter Hogg, Reyer Zwiggelaar

Correlation between Topological Descriptors of the Breast Ductal Network from Clinical Galactograms and Texture Features of Corresponding Mammograms

Mammographic texture has been reported as a biomarker of cancer risk. Recent publications also suggest correlation between the topology of the breast ductal network and risk of cancer. The ductal network can be visualized by galactography, the preferred imaging technique for nipple discharge. We present current results about the correlation between topological and textural properties of clinical breast images. This correlation was assessed for 41 galactograms and 56 mammograms from 13 patients. Topology was characterized using feature extraction techniques arising from text-mining, validated previously in the classification of normal, benign, and malignant galactograms. In addition, we calculated 26 texture descriptors using an automated breast image analysis pipeline. Regression analysis was performed between texture and topological descriptors averaged over all images of the same patient. These data demonstrate a correlation between topology and a subset of texture features with borderline statistical significance due to the limited sample size.

Predrag R. Bakic, David D. Pokrajac, Mathew Thomas, Angeliki Skoura, Tatyana Nuzhnaya, Vasileios Megalooikonomou, Brad Keller, Yuanjie Zheng, Despina Kontos, James C. Gee, Gilda Cardenosa, Andrew D. A. Maidment

Breast Volume Measurement Using a Games Console Input Device

The automated measurement of breast volume has applications both in facilitating the decisions made by surgeons prior to breast reconstruction and in improving density estimation. We describe a novel approach to volume measurement for surgical planning, using a games console input device - the Microsoft Kinect. We have explored the ability of the device to measure surface depth for a range of distances and angles, demonstrating a mean depth error of below 1.5 mm for a distance range of interest (0.5 - 0.8 m). We have also validated the use of the system for volume measurement using a full-sized model female torso. The Kinect-based result is in good agreement with the volume measured by filling a mould of the breast with water (225.5±8.7 ml, 229.4 ±9.7 ml respectively). The method has the potential to provide convenient, cost- and time-effective measurement of breast volume in clinical practice.

Stefanie T. L. Pöhlmann, Jeremy Hewes, Andrew I. Williamson, Jamie C. Sergeant, Alan Hufton, Ashu Gandhi, Christopher J. Taylor, Susan M. Astley

Towards Spatial Correspondence between Specimen and In-vivo Breast Imaging

Radiological in-vivo imaging, such as X-ray mammography and Magnetic Resonance Imaging (MRI), is used for tumour detection, diagnosis and size determination. After tumour excision, histopathological imaging of the stained specimen is used as the gold standard for characterisation of the tumour and surrounding tissue. Relating the information available at the micro and macroscopic scales could lead to a better understanding of the in-vivo radiological imaging. This in turn has potential to improve therapeutic decision making and, ultimately, patient prognosis and treatment outcomes. Accurate alignment of data, necessary to maximise information retrieval from the different scales, can be problematic however, due to the large deformation that the breast tissue undergoes after surgery. In this work we present a methodology to reconstruct a 3D volume from multiple X-ray breast specimen images. The reconstructed volume can be used to bridge the gap between histopathological and in-vivo radiological images. We demonstrate the use of this algorithm on four mastectomy samples. For one of these cases, a specimen MRI was also available and was used to provide an assessment of the performance of the reconstruction technique.

Thomy Mertzanidou, John Hipwell, Mehmet Dalmis, Bram Platel, Jeroen van der Laak, Ritse Mann, Nico Karssemeijer, Peter Bult, David Hawkes

SIFT Texture Description for Understanding Breast Ultrasound Images

Texture is a powerful cue for describing structures that show a high degree of similarity in their image intensity patterns. This paper describes the use of Self-Invariant Feature Transform (SIFT), both as low-level and high-level descriptors, applied to differentiate the tissues present in breast US images. For the low-level texture descriptors case, SIFT descriptors are extracted from a regular grid. The high-level texture descriptor is build as a Bag-of-Features (BoF) of SIFT descriptors. Experimental results are provided showing the validity of the proposed approach for describing the tissues in breast US images.

Joan Massich, Fabrice Meriaudeau, Melcior Sentís, Sergi Ganau, Elsa Pérez, Domenec Puig, Robert Martí, Arnau Oliver, Joan Martí

Comparison of Methods for Current-to-Prior Registration of Breast DCE-MRI

The use of prior studies to complement the information in Breast Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) can help to reduce the currently high false positive ratios. Registration is a fundamental part of this process, as registration algorithms provide automatic correspondences between current and prior studies. The deformable nature of the breast and differences in acquisition protocols make this a particularly challenging problem. In this paper we study three registration algorithms (Affine, SyN and Demons) applied to DCE-MRI images obtained from clinical practice. The methodology followed for this study included using segmentation algorithms in order to focus on the area of the breast. Anatomical landmarks were also added by an expert for evaluation purposes. This allowed us to use an anatomical-landmark-based measure in order to evaluate the quality of registration. Additionally, an image metric was also used for the same purpose. Results, shown to be statistically significant indicate how SyN obtains the best results in terms of the two measures considered.

Yago Diez, Albert Gubern-Mérida, Lei Wang, Susanne Diekmann, Joan Martí, Bram Platel, Johanna Kramme, Robert Martí

A Study on Mammographic Image Modelling and Classification Using Multiple Databases

Within computer aided mammography, there are many image analysis methods have been developed for mammographic image classification. Some of these were developed and validated using well known publicly available databases, and others may have chosen to use independent/private databases for their investigations. Often, despite the promising results described in the literature, it is not unusual to see when adapting an established method with the recommended configurations for a different database, the obtained results are not in line with expectation. This paper presents results of a study with respect to the implications of mammographic image classification using different classifiers trained with variations, such as differences in parameter settings, classifiers, using single databases, combined and across databases. The results indicated that it is unlikely to have an universal parameter settings and classifiers, which can be used to achieve the best classification without tuning. Additional databases used at the training stages do not necessarily lead to more accurate density classifications; whilst classifiers trained with images obtained using one type of image acquisition are not ideal for classifying images obtained using different image acquisition. The related issues of optimal parameter configuration, classifier selection, and utilising single or multiple databases at the training stage are discussed.

Wenda He, Erika R. E. Denton, Reyer Zwiggelaar

Quasi-3D Display of Lesion Locations Simulated by Two Views of Digital Mammography

In the interpretation of digital mammography, intuitive recognition of the spatial location of a lesion projected on the images requires considerable experience for radiologist or radiological technologist. In order to support radiologists and radiological technologists to reading mammography, we have developed a computerized scheme to produce a simulated three-dimensional (3D) display of lesion locations by using craniocaudal (CC) and mediolateral-oblique (MLO) views of digital mammography. In the preliminary results obtained from 20 cases with lesions, 100% of lesions were correctly displayed on the simulated 3D image in which locations were verified by the certificated breast radiological technologist.

Yu Narita, Noritaka Higashi, Yoshikazu Uchiyama, Junji Shiraishi

A Shearlet-Based Filter for Low-Dose Mammography

To improve image quality of low-dose mammography images, we study a new approach of removing Poisson noise from a degraded image in shearlet domain. We first transform Poisson noise into a near Gaussian noise by a shearlet-based multiply variance stabilizing transform (VST). Second, the initial positions of ideal shearlet coefficients are found by thresholding Gaussian noise coefficients. Third, an iterative scheme is proposed to estimate non-noise coefficients from the found initial ideal shearlet coefficients. Finally, the reduced noise image is obtained by the inverse shearlet transform on the estimated coefficients. The main contribution is to combine thresholding and the iterative scheme. A range of experiments demonstrate that the proposed method outperforms the traditional shearlet-based method.

Huiqin Jiang, Yunyi Zhang, Ling Ma, Xiaopeng Yang, Yumin Liu

Evaluation of Human Contrast Sensitivity Functions Used in the Nonprewhitening Model Observer with Eye Filter

Model observers which can serve as surrogates for human observers could be valuable for the assessment of image quality. For this purpose, a good correlation between human and model observer is a prerequisite. The nonprewhitening model observer with eye filter (NPWE) is an example of such a model observer. The eye filter is a mathematical approximation of the human contrast sensitivity function (CSF) and is included to correct for the response of the human eye. In the literature several approximations of the human CSF were found. In this study the relation between human and NPWE observer performance using seven eye filters is evaluated in two-alternative-forced-choice (2-AFC) detection experiments involving disks of varying diameter and signal energy and two background types. The results show that the shape of the CSF has an impact on the correlation between human and model observer. The inclusion of a CSF may indeed improve the relation between human and model observer. However, we did not find an eye filter which is optimal in both backgrounds.

Ramona W. Bouwman, Ruben E. van Engen, David R. Dance, Kenneth C. Young, Wouter J. H. Veldkamp

It Is Hard to See a Needle in a Haystack: Modeling Contrast Masking Effect in a Numerical Observer

Within the framework of a virtual clinical trial for breast imaging, we aim to develop numerical observers that follow the same detection performance trends as those of a typical human observer. In our prior work, we showed that by including spatio-temporal contrast sensitivity function (stCSF) of human visual system (HVS) in a multi-slice channelized Hotelling observer (msCHO), we can correctly predict trends of a typical human observer performance with the viewing parameters of browsing speed, viewing distance and contrast. In this work we further improve our numerical observer by modeling contrast masking. After stCSF, contrast masking is the second most prominent property of HVS and it refers to the fact that the presence of one signal affects the visibility threshold for another signal. Our results indicate that the improved numerical observer better predicts changes in detection performance with background complexity.

Ali R. N. Avanaki, Kathryn S. Espig, Albert Xthona, Tom R. L. Kimpe, Predrag R. Bakic, Andrew D. A. Maidment

Mammography: Radiologist and Image Characteristics That Determine the Accuracy of Breast Cancer Diagnosis

Variations in the performance of breast readers are well reported, but key lesion and reader parameters explaining such variations are not fully explored. This large study aims to: 1) measure diagnostic accuracy of breast radiologists, 2) identify parameters linked to higher levels of performance, and 3) establish the key morphological descriptors that impact detection of breast cancer. Methods: Sixty cases, 20 containing cancer, were shown to 129 radiologists. Each reader was asked to locate any malignancies and provide a confidence rating using a scale of 1-5. Details were obtained from each radiologist regarding experience and training and were correlated with jackknifing free response operating characteristic (JAFROC) figure of merit. Cancers were ranked according to the “detectability rating” that is, the number of readers who accurately detected and located the lesion divided by the total number of readers, and this was correlated with various mathematical lesion descriptors. Results: Higher reader performance was positively correlated with number of years reading mammograms (r=0.24, p=0.01), number of mammogram readings per year (r=0.28, p=0.001), and hours reading mammogram per week (r=0.19, p=0.04). For image features and lesion descriptors there was correlation between “detectability rating” and lesion size (r=0.65, p=0.005), breast density (r=-0.64, p=0.007), perimeter (r=0.66, p=0.0004), eccentricity (r= 0.49, p=0.02), and solidity (r=0.78, p< 0.0001). Radiologist experience and lesion morphology may contribute significantly to reduce cancer detection.

Mohammad A. Rawashdeh, Claudia Mello-Thoms, Roger Bourne, Patrick C. Brennan

Preliminary Study on Sub-Pixel Rendering for Mammography Medical Grade Color Displays

The independent sub-pixel driving (ISD) technology, which utilizes sub-pixels included in each normal pixel for image rendering, had been developed for monochrome displays for mammography to improve their resolution and over–all noise properties, and displays with ISD technology has been used for the diagnostic reading on mammography in many institutions. The purpose of this preliminary study was to investigate a possibility of applying the ISD technology to medical color displays on mammography using quantitative resolution and noise evaluations and a visual comparison. A prototype 5 mega-pixel (MP) color display and a 5MP monochrome display were employed. To reduce the micro level color shifts caused by applying the existing ISD driver software to color displays, we implemented an additional low-pass filtering process to the driver software. The quantitative over-all resolution and noise properties for a magnification ratio of 0.4 which is routinely used in diagnostic initial reading of our hospital was measured for three conditions which included two conditions for the color display: Color with ISD (Color-ISD) and color without ISD (Color-normal) and one condition for the monochrome display: Monochrome-normal. Two radiologist visually compared Color-ISD with Monochrome-normal for resolution, noise and color shift using an ACR156 phantom image. Both over-all resolution and noise properties of ISD-color were superior to those of the others. In the visual comparison, Color ISD presented the similar resolution and superior noise properties as compared with Monochrome-normal. The color shift was visually ignorable in the phantom image displaying.

Katsuhiro Ichikawa, Hiroko Kawashima

Impact of Color Calibration on Breast Biopsy Whole Slide Image Interpretation Accuracy and Efficiency

Color LCD use is increasing in medical imaging especially in applications like telepathology. Standardized methods for calibrating, characterizing and profiling color displays have not been created. We used a validated calibration, characterization and profiling protocol for color medical imaging applications to determine if it impacts performance accuracy and interpretation time. 250 breast biopsy whole slide image (WSI) areas (half malignant, half benign) were displayed to 6 pathologists. In one condition the calibration protocol was used and in the other the same display was un-calibrated. Receiver Operating Characteristic area under the curve (Az) with the calibrated display was 0.8570 and with the un-calibrated one was 0.8488 (p = 0.4112). For interpretation time, the mean with the calibrated display was 4.895 sec and with the un-calibrated display was 6.304 sec (p = 0.0460). There is an advantage diagnostically using a properly calibrated and color-managed display and a significant advantage for potentially improving workflow via reduced viewing times.

Elizabeth A. Krupinski, Louis D. Silverstein, Syed F. Hashmi, Anna R. Graham, Ronald S. Weinstein, Hans Roehrig

Backmatter

Weitere Informationen

Premium Partner

    Bildnachweise