Skip to main content
Top

2018 | Book

Bildverarbeitung für die Medizin 2018

Algorithmen - Systeme - Anwendungen. Proceedings des Workshops vom 11. bis 13. März 2018 in Erlangen

Editors: Prof. Dr. Andreas Maier, Prof. Dr. Thomas M. Deserno, Prof. Dr. Heinz Handels, Dr. Klaus Hermann Maier-Hein, Prof. Dr. Christoph Palm, Prof. Dr. Thomas Tolxdorff

Publisher: Springer Berlin Heidelberg

Book Series : Informatik aktuell

insite
SEARCH

About this book

In den letzten Jahren hat sich der Workshop "Bildverarbeitung für die Medizin" durch erfolgreiche Veranstaltungen etabliert. Ziel ist auch 2018 wieder die Darstellung aktueller Forschungsergebnisse und die Vertiefung der Gespräche zwischen Wissenschaftlern, Industrie und Anwendern. Die Beiträge dieses Bandes - einige davon in englischer Sprache - umfassen alle Bereiche der medizinischen Bildverarbeitung, insbesondere Bildgebung und -akquisition, Maschinelles Lernen, Bildsegmentierung und Bildanalyse, Visualisierung und Animation, Zeitreihenanalyse, Computerunterstützte Diagnose, Biomechanische Modellierung, Validierung und Qualitätssicherung, Bildverarbeitung in der Telemedizin u.v.m.

Table of Contents

Frontmatter
Systems Medicine
The Next Generation of Computer-assisted Precision Medicine

Recent advances in modern OMICS technology allow measuring the expression of all kinds of biological entities (genes, proteins, metabolites, miRNAs, etc.) at low cost and in high-throughput. Computational challenges for analyzing such big data emerge, ranging from the low signal to noise ratio to high model complexity, which render simple statistical questions arbitrarily complicated.We will discuss several bioinformatics tools for de-isolating biological networks and multiple OMICS data types: de novo pathway enrichment, in vitro high-throughput screening (HTS) data integration, time-course network enrichment, cancer subtyping, and breath analysis.

Jan Baumbach
Reinventing Bone Surgery
From Planning to Execution of a Hard-Tissue Cut

Cutting bones is one of the oldest medical procedures performed to human patients. Thanks to the high mineral content of bone we have archaeological evidence of skull trepanation dating back more than 10.000 years. Despite the rapid development of surgical instruments over the last 200 years, the fundamental mechanism of bone cutting has not changed ever since.

Philippe Cattin
From Mechanistic to Data-driven Models for Surgical Planning, Guidance and Simulation

Biomechanical and biophysical models are key tools in many applications of surgical planning and optimisation, surgical guidance, and interactive simulation for training and rehearsal. The most robust and accurate models usually are those based on the relevant equations of continuum mechanics (solid, fluid, thermal, etc.), and which are generally solved with numerical methods such as FEM. Given high quality patient-specific inputs, these can enable accurate prediction of, e.g., deformations of soft tissues, flow patterns in blood vessels, energy delivery profiles around ablation devices, etc.

Zeike A. Taylor
Precision Imaging
From Population Imaging Analytics to In-silico Clinical Trials

Medical image computing is witnessing exciting times. Specifically to this talk, new opportunities and challenges have emerged with the growing availability of large population imaging repositories being collected in the UK, USA, Canada, Germany, and The Netherlands, to name Just a few.

Alejandro Frangi
Deep Learning Fundamentals

Deep learning has received a lot of attention in the machine learning community. Successful applications from speech recognition or computer vision are already part of our daily life. Much effort has been devoted to transferring this success to medical image computing.

Katharina Breininger, Vincent Christlein, Tobias Würfl, Andreas Maier
Advanced Deep Learning Methods

The remarkable rise of deep learning has led to an overwhelming amount of new papers coming up by the week. This tutorial intents to filter out the research most relevant for the medical image computing (MIC) community and present it in a structured and understandable form. It is composed of five parts: Classification, Segmentation, Detection, Generative Models and Semi- Supervised Learning.

Paul Jäger, Fabian Isensee, Jens Petersen, David Zimmerer, Jakob Wasserthal, Klaus Hermann Maier-Hein
Innovation Generation, Disruption and Exponential Technologies in Medical Imaging

Healthcare in general and medical imaging in particular will change dramatically in the next 20 years with the emergence of artificial intelligence (including machine and deep learning), 3D printing, personalized diagnosis and therapies, shift from therapy to prevention, and many related delivery and economic changes. Huge opportunities are present in developed, emerging and developing nations, each with completely different needs and infrastructural environments, and with a completely different reimbursement system. The tutorial will prepare the attendees for the changes that will likely come up in the coming years and subsequently focus on recognizing the need for a change in our development and innovation process, that must focus on digitization and connectivity, small footprint, robustness, and low cost, and with that explore alternative financing methods.

Michael Friebe
Endoscopy

In the last 40 years progress in the field of diagnostic and interventional endoscopy (gastroenterology and laparoscopy) have been made, mainly driven through new technical developments from various areas of medical technologies such as optics, micro-mechanics, electronics, robotics, informatics, and image processing. These advances allow soft and gentle minimal-invasive access and interventions in all hollow organs, including the abdomen, stomach, large and small intestines, joints a was well nasal cavities. To accelerate the development and application of such diagnostic and therapeutic possibilities in all fields of endoscopy a understanding of the available technological possibilities is necessary.

Thomas Wittenberg
Abstract: Digitale Pathologie für mobile Endgeräte

Digitalisierung spielt im heutigen Zeitalter, besonders in der Medizin, eine entscheidende Rolle. Neben der Radiologie, welche bereits digitalisiert ist, gibt es auch in anderen Fachdisziplinen eine Umstellung bisheriger Workflows. So hat sich auch in den letzten Jahren die digitale Pathologie weiterentwickelt.

Hannah Büchner, Ingmar Gergel
Abstract: Digital Cytomorphology
Deep Learning on an Image Data Set of Cell Morphologies in Acute Myeloid Leukemia

Examination of Leukocyte cytomorphology using light microscopy, a method dating back to the nineteenth century, remains an important cornerstone in present-day Leukemia diagnostics.

Christian Matek, Carsten Marr, Karsten Spiekermann
Abstract: First Approaches Towards Automatic Detection of Microaneurysms in OCTA Images

We investigated automatic detection of micro aneurysms in optical coherence tomo-graphy angiography. Data of two patients was gathered at the New England Eye Center. Patients with diabetic retinopathy were imaged on an Optovue Avanti device.

Lennart Husvogt, A. Yasin Alibhai, Eric Moult, James G. Fujimoto, Nadia Waheed, Andreas Maier
Abstract: Automatic Malignancy Estimation for Pulmonary Nodules from CT Images

Early detection of lung cancer is crucial to increase the chance of cure. As lung cancer often manifests itself in the presence of malignant pulmonary nodules, the assessment of such is of high clinical importance. Lung cancer screening is primarily performed using diagnostic imaging modalities such as CT, while invasive methods such as biopsy are used as a last resort to confirm diagnosis.

Katrin Mentl, Rimon Saffoury, Andreas Maier
Abstract: Amplitude of Brain Signals Classify Hunger Status based on Machine Learning in Resting-state fMRI

Resting-state fMRI (rs-fMRI) allows for a task-free exploration of the human brain’s intrinsic functional connectivity. Since central nervous pathways regulate food intake and eating behavior, it is assumed that changes in the homeostatic state have an impact on the connectivity patterns of rs-fMRI. Here, we compare the accuracy of three data-driven approaches in classifying two metabolic states (hunger vs satiety) depending on the observed rs-fMRI fluctuations.

Arkan Al-Zubaidi, Alfred Mertins, Marcus Heldmann, Kamila Jauch-Chara, Thomas F. Münte
Abstract: Assessment of Segmentation Dependence in Macroscopic Lung Cavity Extraction

Training of respiratory motion models and population-based patient phantoms of the lung often requires the definition of the entire lung cavity region in the 4D-CT. To ease the workload of clinical experts, automatic selection is highly desirable. Many lung cavity extraction methods rely on a pre-segmented lung volume.

Asmaa Khdeir, Tobias Geimer, Shuqing Chen, Eric Goppert, Maximilian Dankbar, Christoph Bert, Andreas Maier
Abstract: Percutaneous Pelvis Fixation Using the Camera-augmented C-arm
First Successes in Ex Vivo Deployment

Today, percutaneous techniques are widely accepted for treatment of bone fractures in spine and pelvis. These techniques are enabled by modern imaging technology, such as mobile C-arm X-ray machines, and allow for substantial reductions in blood loss, collateral tissue damage, and overall surgery duration [1]. While minimally invasive surgery is beneficial for the patient, it increases the task load for the surgeon.

Mathias Unberath, Javad Fotouhi, Emerson Tucker, Alex Johnson, Greg Osgood, Nassir Navab
Abstract: Augmented Reality im Operationssaal

Die Integration präoperativer Bilddaten in den Arbeitsablauf im Operationssaal ist eine Herausforderung, die mit immer leistungsfähigeren Bildgebungsverfahren an Dringlichkeit gewinnt. Sie steht unter der Randbedingung, Bildinformationen kontextspezifisch ohne Störung des Eingriffs bereitzustellen.

Stephan Vedder
Abstract: Efficient Labeling of Optical Coherence Tomography Angiography Data using Eye Tracking

We implemented an approach for the efficient labeling of structural and angiographic optical coherence tomography (OCT) data using eye tracking. OCT is a non-invasive imaging technology, which provides volumetric data from scattering tissues in micrometer resolution. Due to its widespread use and noninvasiveness, clinicians acquire large amounts of volumetric ophthalmic data on a daily basis.

Lennart Husvogt, Eric Moult, Nadia Waheed, James G. Fujimoto, Andreas Maier
Abstract: Leveraging Open Source Software to Close Translational Gaps in Medical Image Computing

Many imaging biomarkers (IBs) fail clinical translation. The main reason is not a lack of utility, but translational gaps [1] during validation and qualification. One important problem in this context is the landscape of existing IT systems in the clinical environment. Systems are highly heterogeneous and proprietary, causing significant translational challenges that are often purely infrastructural in nature.

Jens Petersen, Sabine Heiland, Marti Bendszus, Jürgen Debus, Marco Nolden, Caspar J. Goch, Klaus Hermann Maier-Hein
3D-CNNs for Deep Binary Descriptor Learning in Medical Volume Data

Deep convolutional neural networks achieve impressive results in many computer vision tasks not least because of their representation learning abilities. The translation of these findings to the medical domain with large volumetric data e.g. CT scans with typically ≥ 106 voxels is an important area of research. In particular for medical image registration, a standard analysis task, the supervised learning of expressive regional representations based on local greyvalue information is of importance to define a similarity metric. By providing discriminant binary features modern architectures can leverage special operations to compute hamming distance based similarity metrics. In this contribution we devise a 3D-Convolutional Neural Network (CNN) that can efficiently extract binary descriptors for Hamming distance-based metrics. We adopt the recently introduced Binary Tree Architectures and train a model using paired data with known correspondences. We employ a triplet objective term and extend the hinge loss with additional penalties for non-binary entries. The learned descriptors are shown to outperform state-of-the-art hand-crafted features on challenging COPD 3D-CT datasets and demonstrate their robustness for retrieval tasks under compression factors of ≈ 2000.

Max Blendowski, Mattias P. Heinrich
Detecting and Measuring Surface Area of Skin Lesions

The treatment of skin lesions of various kinds is a common task in clinical routine. Apart from wound care, the assessment of treatment efficacy plays an important role. Fully manual measurements and documentation of the healing process can be very cumbersome and imprecise. Existing technical solutions often require the user to delineate the lesion manually and rarely provide information on measurement precision or accuracy. We propose a method for segmenting and measuring lesions using a single image. Surface area of lesions on bent surfaces is estimated based on a paper ruler. Only roughly outlining the region of interest is required. Wound segmention evaluation was performed on 10 images, resulting in an accuracy of 0.98 ± 0.02. For surface measuring evaluation on 40 phantom images we found an absolute error of 0.32 ± 0.27 cm2 and a relative error of 5.2 ± 4.3%.

Houman Mirzaalian-Dastjerdi, Dominique Töpfer, Michael Bangemann, Andreas Maier
Abstract: Deep Hashing for Large-Scale Medical Image Retrieval

Adoption of content-based image retrieval systems (CBIR) requires efficient indexing of the data contents in order to respond to visual queries without explicitly relying on textual keywords. Searching for similar data is closely related to the fundamental problem of nearest neighbor search. Exhaustive comparison of a query across the database is infeasible in large-scale retrieval as it is computationally expensive [1].

Sailesh Conjeti, Magdalini Paschali, Abhijit Guha Roy, Nassir Navab
Abstract: Physiological Parameter Estimation from Multispectral Images Unleashed

Multispectral imaging in laparoscopy can provide tissue reflectance measurements for each point in the image at multiple wavelengths of light. These reflectances encode information on important physiological parameters not visible to the naked eye. Fast decoding of the data during surgery, however, remains challenging.

Sebastian J. Wirkert, Anant S. Vemuri, Hannes G. Kenngott, Sara Moccia, Michael Götz, Benjamin F. B. Mayer, Klaus Hermann Maier-Hein, Daniel S. Elson, Lena Maier-Hein
Abstract: Probabilistic Appearance Models for Medical Image Analysis

The identification of one-to-one point correspondences between image objects is one key aspect and at the same time the most challenging part of generating statistical shape and appearance models. Using probabilistic correspondences between samples instead of accurately placed landmarks for shape models [1] eliminated the need of extensive and time consuming landmark and correspondence determination, and furthermore, the dependency of the quality of the generated model on potentially wrong correspondences was reduced.

Julia Krüger, Jan Ehrhardt, Heinz Handels
Abstract: Robust Multi-Scale Anatomical Landmark Detection in Incomplete 3D-CT Data

An essential prerequisite for comprehensive medical image analysis is the robust and fast detection of anatomical structures in the human body. To this point, machine learning techniques are most often applied to address this problem, exploiting large annotated image databases to estimate parametric models for anatomy appearance. However, the performance of these methods is generally limited, due to suboptimal and exhaustive search strategies applied on large volumetric image data, e.g., 3D-CT scans.

Florin C. Ghesu, Bogdan Georgescu, Sasa Grbic, Andreas Maier, Joachim Hornegger, Dorin Comaniciu
Abstract: Exploring Sparsity in CNNs for Medical Image Segmentation BRIEFnet

Deep convolutional neural networks can evidently achieve astonishing accuracies for multiple medical image analysis tasks, in particular segmentation and detection. However, the actual translation of deep learning into clinical practice is so far very limited, in part because their extensive computations rely on specialised GPU hardware that is not easily available in clinical environments.

Mattias P. Heinrich, Ozan Oktay
Abstract: Fast MRI Whole Brain Segmentation with Fully Convolutional Neural Networks

Whole brain segmentation from structural MRI-T1 scan is a prerequisite for most morphological analyses, but requires hours of processing time and therefore delays the availability of image markers after scan acquisition. We introduced a fully convolution neural network (F-CNN) that segments a brain scan in several seconds [1]. Training deep F-CNNs for semantic image segmentation requires access to abundant labeled data.

Abhijit Guha Roy, Sailesh Conjeti, Nassir Navab, Christian Wachinger
Patient Surface Model and Internal Anatomical Landmarks Embedding

The patient surface model has shown to be a useful asset to improve existing diagnostic and interventional tasks in a clinical environment. For example, in combination with RGB-D cameras, a patient surface model can be used to automate and accelerate the diagnostic imaging workflow, manage patient dose, and provide navigation assistance. A shortcoming of today’s patient surface models, however, is that, internal anatomical landmarks are not present. In this paper, we introduce a method to estimate internal anatomical landmarks based on the surface model of a patient. Our method relies on two major steps. First, we fit a template surface model is to a segmented surface of a CT dataset with annotated internal landmarks using keypoint and feature descriptor based rigid alignment and atlas-based non-rigid registration. In a second step, we find for each internal landmark a neighborhood on the template surface and learn a generalized linear embedding between neighboring surface vertices in the template and the internal landmark. We trained and evaluated our method using cross-validation in 20 datasets over 50 internal landmarks. We compared the performance of four different generalized linear models. The best mean estimation error over all the landmarks was achieved using the lasso regression method with a mean error of 12.19 ± 6.98 mm.

Xia Zhong, Norbert Strobel, Annette Birkhold, Markus Kowarschik, Rebecca Fahrig, Andreas Maier
Comparison of Self-similarity Measures for Multi-modal Non-rigid Registration of 3D-PLI Brain Images

We introduce self-similarity measures in a spline-based nonrigid registration method. We applied our method to register multimodal 3D polarized light imaging and blockface image data of human and rat brain sections. Quantitative evaluations demonstrate that using self-similarity measures increases the accuracy and robustness compared to a traditional mutual information measure.

Sharib Ali, Dehui Lin, Markus Axer, Karl Rohr
Two-Step Trajectory Visualization for Robot-Assisted Spine Radiofrequency Ablations

Radiofrequency Ablations (RFAs) can be employed for the treatment of spine metastases. Instruments are therefor inserted through the vertebra’s pedicle into cancerous tissue within the vertebral body. This requires high precision during interventions. We present a two-step method to increase risk awareness during intervention planning and execution of manual and robot-assisted spine RFAs. Three medical experts evaluated our method and stated that it yields two advantages: First, improved visualizations for manual interventions and second, increased safety in hand-guided, robot-assisted setups.

Nico Merten, Simon Adler, Magnus Hanses, Sylvia Saalfeld, Mathias Becker, Bernhard Preim
Unsupervised Pathology Detection in Medical Images using Learning-based Methods

Detecting pathologies automatically is challenging because of their big variability. As the usual supervised machine learning approaches would only be able to detect one type of pathologies, in this work we pursue an unsupervised approach: learn the entire variability of healthy data and detect pathologies by their differences to the learned norm. Two methods have been developed based on this principle: A modified PatchMatch algorithm shows plausible results on contrasting brain tumors, but bad generalization ability for other types of data. A CVAE-based method on the other hand performs significantly better and ca. 17 times faster on the brain data and can be generalized to other pathologies, e.g. lung tumors. Not only is the achieved Dice coefficient of 0.55 comparable to other supervised methods on this data, moreover this method reliably detects different pathology types and needs no groundtruth.

Hristina Uzunova, Heinz Handels, Jan Ehrhardt
Classification of Lobular and Ductal Breast Carcinomas by Texture Analysis in DCE-MRI Data

Breast cancer can be distinguished into several subtypes, where invasive ductal carcinomas (IDC) and invasive lobular carcinomas (ILC) are the two most common subtypes. These two types of tumor grow at a different speed and exhibit different metastatic patterns. Although both types are malignant, they show different treatment results for the same therapy. Accurate distinction between these two subtypes is helpful for determining therapy strategies. In this paper, we classify IDC and ILC based on their characteristic texture features, which are extracted from a three-dimensional co-occurrence matrix. The texture features at different time points are used instead of the features from a single time point. We employ a non-linear support vector machine (SVM) algorithm and a random forests method as classifiers to separate IDC and ILC via their texture features and achieve a high accuracy of the classification result.

Kai Nie, Gabriel Mistelbauer, Bernhard Preim
Abstract: Revealing Hidden Potentials of the q-Space Signal in Breast Cancer

Mammography screening for early detection of breast lesions currently suffers from high amounts of false positive findings, which result in unnecessary invasive biopsies. Diffusion-weighted MR images (DWI) can help to reduce many of these false-positive findings prior to biopsy. Current approaches estimate tissue properties by means of quantitative parameters taken from generative, biophysical models fit to the q-space signal under certain assumptions regarding noise and spatial homogeneity. This process is prone to fitting instability and partial information loss due to model simplicity.

Paul Jaeger, Sebastian Bickelhaupt, Frederik Bernd Laun, Wolfgang Lederer, Daniel Heidi, Tristan Anselm Kuder, Daniel Paech, David Bonekamp, Alexander Radbruch, Stefan Delorme, Heinz-Peter Schlemmer, Franziska Steudle, Klaus Hermann Maier-Hein
Elastic Mitral Valve Silicone Replica Made from 3D-Printable Molds Offer Advanced Surgical Training

Reconstructive mitral valve surgeries are demanding cardiac surgeries that are conducted in the mid-age of a surgical career at the earliest. Receiving years of training in patients is required to gain the experiences and skills necessary for surgical success. On top of that, the number of such surgeries is limited per hospital, therefore other means of training should be offered to facilitate a steep learning curve. Within the scope of this work, we equipped an existing physical simulator with patient-specific flexible replica of the mitral valve. We developed software to automatically produce a 3D-printable casting mold for silicone material, as this mimics properties of the valve tissue in terms of stitching and cutting. We show the feasibility of the approach and the usefulness of these models is evaluated by an experienced cardiac surgeon, successfully conducting reconstructive surgery on pathological silicone valves.

Sandy Engelhardt, Simon Sauerzapf, Sameer Al-Maisary, Matthias Karck, Bernhard Preim, Ivo Wolf, Raffaele De Simone
Simulation of Realistic Low Dose Fluoroscopic Images from their High Dose Counterparts

Learning based denoising methods have attracted increasing interest in the recent past. These methods rely on data pairs. In the case of denoising, the data pairs are usually noise corrupted images and their noise-free counterparts, if available. Otherwise an associated high-dose X-ray image can be used instead as a practical alternative. As the current image processing techniques are not yet able to provide the necessary image quality at very low dose levels, it is usually not possible to acquire clinical sequences. As variation in the data is extremely important for learning based methods, phantom data alone cannot be used to train a network and achieve optimal performance. A possibility to overcome this issue is to simulate low dose images from the related high dose images. However, to make sure that the simulated low dose images are realistic (replicate the properties of real low dose images), image noise attributes associated with low dose image acquisition need to be taken into account. In their paper we introduce a novel method to simulate low dose images from high dose images based on modelling the X-ray image formation process. This way, we can better account for imaging parameters such as system gain and electronic noise. We have evaluated our method by comparing several corresponding regions of the simulated lower dose images with that of real lower dose images using a two sample Kolmogorov-Smirnov Test at 5% significance. Out of 40 pairs, in 85% of the cases the hypothesis that the corresponding regions (from the low and simulated low dose images) belong to the same distribution has been accepted.

Sai Gokul Hariharan, Norbert Strobel, Markus Kowarschik, Rebecca Fahrig, Nassir Navab
Towards Full-body X-ray Images

Digital tomosynthesis is a tomographic imaging technique whose upsurge is mainly caused by breast imaging. However, it might also be useful in orthopedics due to its high in-plane resolution as well as the fact that tomosynthetic slices do not suffer from magnification or distortion, making measurements possible, for example, even without the need of any calibration object. Since the reading time of such a reconstruction is higher compared to conventional 2-D radiographs, a simple parallel projection of the volume can be computed to get an overview of the volume. However, this leads to a rather blurred image impression since all artifacts and inhomogeneities in the reconstructed volume as well as certain anatomical structures which are not necessary for the diagnosis, will end up in the projection. We propose a method which selects the slices to be projected into a smart synthetic X-ray image in a way which is optimal w.r.t to the sharpness of predefined ROIs (e. g. knee, spine or hip). Therefore, two Laplacian-based auto-focus measures are combined with a thin-plate spline yielding a sharp and homogenous image impression within the smart radiograph. It was shown that the auto-focus method is able to select the same slice as have been selected during an expert annotation. Upon visual inspection, it could be determined that the proposed method achieves higher contrast and clearly better visibility of complex bone structures like spine or hip.

Christoph Luckner, Thomas Mertelmeier, Andreas Maier, Ludwig Ritschl
Influence of Excitation Signal Coupling on Reconstructed Images in Magnetic Particle Imaging

In Magnetic Particle Imaging, superparamagnetic iron oxide nanoparticles are excited by an oscillating magnetic field that is generated by sending coils. In multidimensional set-ups, perfect geometrical decoupling cannot be achieved. Remaining coupling can distort the trajectory and introduce artefacts, thus active decoupling is introduced at costs of high power consumption. This work uses a device with active coupling control to investigate the influence of different coupling levels on the image reconstruction.

Anselm von Gladiss, Matthias Graeser, Thorsten M. Buzug
A Joint Probabilistic Model for Speckle Variance, Amplitude Decorrelation and Interframe Variance (IFV) Optical Coherence Tomography Angiography

Optical Coherence Tomography Angiography (OCTA) is a general method to visualize blood flow in biological tissue. Despite its good results in practice, the commonly used Amplitude Decorrelation OCTA (AD-OCTA) measure suffers from a well-understood objective function, which makes it challenging to mathematically model post processing tasks like, e.g., denoising. In this paper, a probabilistic model is developed for the three OCTA measures Speckle Variance OCTA, ADOCTA and the newly proposed Interframe Variance OCTA (IFV-OCTA) to enable further tasks like regularization-based denoising. From a theoretical point of view, IFV-OCTA is shown to be in-between the other two methods and can act as a link between them. A small sized observer study suggests that the image quality of IFV-OCTA is comparable to the other methods. IFV-OCTA is a promising OCTA measure for algorithms that require a dependency on the interscan time.

Stefan B. Ploner, Christian Riess, Julia Schottenhamml, Eric M. Moult, Nadia K. Waheed, James G. Fujimoto, Andreas Maier
A Simulation Study and Experimental Verification of Hand-Eye-Calibration using Monocular X-Ray

In this paper, the simultaneous hand-eye/robot-world problem AX = ZB is performed using a single X-ray image instead of a stereo camera in order to avoid the additional tracking device. Our setup consists of a special X-ray marker, several image preprocessing steps, and a monocular pose estimation algorithm, for extracting the 6- D pose of the marker with respect to the X-ray source. Simulations are performed to investigate the behavior of the proposed hand-eye method when including inaccuracies of the robot and the non-isotropic errors of monocular pose estimation. The simulations were evaluated in an experimental setup, reaching an accuracy of 0.06◦ and 0.77mm.

Petra Dorn, Peter Fischer, Holger Mӧnnich, Philip Mewes, Muhammad Asim Khalil, Abhinav Gulhar, Andreas Maier
Background Correction and Stitching of Histological Plaque Images

Histological examination of atherosclerotic plaques is the gold standard for the analysis of vessel plaque composition. The digitalization of the microscopic histology images results in a set of image tiles with overlapping regions of the same histological 2D slice. To allow comparison with other imaging modalities the tiles must be stitched together to a complete image of the plaque. The purpose of this work is to develop custom processing methods for the intensity correction and stitching problems. The developed methods are applied to 19 plaque images from an ongoing study. Results are compared with manual as well as automatic photo stitching.

Lilli Kaufhold, Heike Goebel, Hanieh Mirzaee, Christoph Strecker, Andreas Harloff, Anja Hennemuth
Towards In-Vivo X-Ray Nanoscopy
The Effect of Motion on Image Quality

Novel X-Ray Microscopy (XRM) systems allow to study the internal structure of a specimen on nanoscale. A possible use of this non-destructive technology is motivated in the medical research area. In-Vivo investigation of medication over a period of time and its effects on perfusion and bony structure might lead to a better understanding of drug mechanisms and diseases like Osteoporosis and could lead to new approaches to their treatment. The first step towards in-vivo XRM imaging is to investigate the suitability of recent XRM systems for this task and subsequently to determine the system parameters. In this context, the impact of mice motion on the image quality is studied in this work. This paper aims to simulate the effects of breathing motion and muscle relaxation of the mice on the reconstructed images, which already effects the projection images. We therefore assume a mouse’s respiration motion pattern, which happens four time during a single projection acquisitions, and the muscle relaxation movement due to anesthesia and simulate its impacts on image quality. Additionally, we show that a frame rate of at least 16 fps is needed to capture in-vivo movements in order to apply state-of-the-art motion correction methods.

Leonid Mill, Bastian Bier, Christopher Syben, Lasse Kling, Anika Klingberg, Silke Christiansen, Georg Schett, Andreas Maier
Towards Fully Automated Determination of Laryngeal Adductor Reflex Latencies through High-Speed Laryngoscopy Image Processing

Protective reflexes of the larynx help to avoid intrusion of foreign particles into the lower airways, which can lead to aspiration pneumonia. These protective mechanisms include the Laryngeal Adductor Reflex (LAR), a rapid adduction of the vocal folds. Up to now, the LAR latency could only be determined manually by visually assessing laryngoscopic high-speed video sequences obtained during and after stimulation of the larynx by water droplet impact. Here, we present a novel image processing algorithm based on difference image calculation and optical flow analysis for a more objective LAR latency determination. To evaluate our prototype algorithm, we compared the results obtained for a set of example sequences with the values given by two expert phoniatricians. The results show a very good LAR stimulation detection performance. LAR onset detection remains challenging for our algorithmic approach as well as for the human perceptive system, as demonstrated by a low inter-rater reliability.

Jacob F. Fast, Martin Ptok, Michael Jungheim, Robin Szymanski, Tobias Ortmaier, Lüder A. Kahrs
Fourier-based Reduction of Directed Streak Artifacts in Cone-Beam CT

Due to its adjustable scan trajectory, C-arm cone-beam CT has been used recently to acquire knee scans in an upright position. However, stabilization devices located outside the FOV introduce streak artifacts in the reconstructed images. This paper proposes a method to remove those streak artifacts. Using selective filtering of the Fourier transforms of the reconstructions, we propose a filter design that attenuates the frequencies that are responsible for the streak artifacts. The filter is constructed by taking both the frequency and the orientation of the introduced streaks into account. We compare our approach to a bandpass-filter. Our proposed method is able to reduce the streaks in the reconstruction remarkably while preserving edge information, whereas the bandpass-filter is not capable of preserving sharp edges in the filtered image. Moreover, our method yields an improved SSIM when comparing both filter techniques to simulated ground truth data.

Julia Gawellek, Bastian Bier, Garry Gold, Andreas Maier
An Open Source Tool for Creating Model Files for Virtual Volume Rendering in PDF Documents

Volume rendering is an important technique for medical imaging where many modalities produce three-dimensional (3D) images. An appropriate three-dimensional rendering leads to a better perception of the image content. A major problem is exchangeability: Usually, only two-dimensional, static snapshots of a volume-rendered scene can be distributed electronically. The Portable Document Format (PDF) provides the possibility to embed 3D objects. With suitable reading software, these objects can be displayed interactively. This article presents an open-source implementation of a software tool that is based on the MeVisLab imaging framework and that can convert volume images into model files which can be embedded into PDF files to create a virtual volume rendering.

Axel Newe, Julian Brandner, Wolfgang Aichinger, Linda Becker
Comparison of Divergence-Free Filters for Cardiac 4D PC-MRI Data

4D PC-MRI enables the measurement of time-resolved blood flow directions within a 3D volume. These data facilitate a comprehensive qualitative and quantitative analysis. However, noise is introduced, e.g., due to inhomogeneous magnetic field gradients. Blood is commonly assumed as a non-Newtonian fluid, thus, incompressible, and divergence should be zero. Divergence-free filters enforce this model assumption and have been shown to improve data quality. In this paper, we compare binomial smoothing and three of these techniques: The finite difference method (FDM), divergence-free radial basis functions (DFRBF) and divergence-free wavelets (DFW). The results show that average and maximum velocities tend to decrease, while average line lengths tend to increase slightly. We recommend FDM or DFW divergence-free filtering as an optional pre-processing step in 4D PC-MRI processing pipelines, as they have feasible computation times of few seconds.

Mickäel Francisco Sereno, Benjamin Köhler, Bernhard Preim
Employing Spatial Indexing for Flexibility and Scalability in Brain Biopsy Planning

Planning of deep brain tumor biopsy is a time intensive task and the result highly dependent on tumor position and patient individual anatomy. The decision on the best needle trajectory is generally based on expert knowledge on optimal entry points and angles as well as trajectory length and rigid rules in respect to avoidance of and safety margins to risk structures. The increasing availability of more detailed data on brain anatomy further increases the complexity of the planning task. However, current computer supported planning systems generally work with fixed rules and a limited set of structures at risk. We propose BrainXPlore, a visual analytics based planning tool allowing neurosurgeon to interactively explore and refine the space of possible trajectories in the context of different quality measures and to define custom rules. To ensure interactivity and performance even for a high number of anatomical structures, we employ a spatial index allowing to access distance information for trajectories in real time. We evaluated BrainXPlore on real brain biopsy planning tasks and conclude that our system can decrease the time needed for biopsy planning and aid novice users in their decision-making process.

Lukas Pezenka, Stefan Wolfsberger, Katja Bühler
Measuring Finger Lengths from 2D Palm Scans

A goal of Life Child is to study the development of children and adolescents. The growth of fingers and other palm compartments in this age group has been received little attraction so far. Usually, finger lengths are measured manually even when 2D palm images have been produced. This is often cumbersome for very large studies. In this paper, we introduce an approach to automatically segement palm and finger compartments of scanned 2D palm scans. The scans were taken by a single document scanner with the goal to measure finger lengths. Our algorithms are rotation invariant, automatically recognize hand objects in images using a skin color model, determine the finger segments for that the length from the fingertip to the crease is derived. We outline steps of the image processing pipeline and show first evaluation results

Alexander Twrdik, Ulf-Dietrich Braumann, Franziska Abicht, Wieland Kiess, Toralf Kirsten
Towards Analysis of Mental Stress Using Thermal Infrared Tomography

A number of publications has focused on detecting and measuring mental stress using infrared tomography as it is a noninvasive and convenient monitoring method. Several potential facial regions of interest such as forehead, nose and the upper lip in which stress may potentially be detectable have been identified in previous contributions. However, these publications are not comparable since they all rely on different approaches regarding both experiment design (stressor, ground truth/reference measurements) as well as evaluation methodology such as either average temperature monitoring or advanced image processing methods. We therefore focus on two aspects: Designing an experiment that allows a reliable induction of mental stress and measuring temperature changes in all aforementioned regions as well as on introducing and evaluating a GLCM-based method for quantitative analysis of the recorded image data. We show that signals extracted from the upper lip region correspond well with high stress levels, while no correspondence can be shown for the other regions. The suggested GLCM-based method is shown to be more specific towards stress response than established measurements based on average region temperature.

Marcin Kopaczka, Thomas Jantos, Dorit Merhof
Preliminary Study Investigating Brain Shift Compensation using 3D CBCT Cerebral Vascular Images

During a neurosurgical procedure, the exposed brain undergoes an elastic deformation caused by numerous factors. This deformation, also known as brain shift, greatly affects the accuracy of neuronavigation systems. Non-rigid registration methods based on point matching algorithms are frequently used to compensate for intraoperative brain shift, especially when anatomical structures such as cerebral vascular tree are available. In this work, we introduce a pipeline to compensate for the volumetric brain deformation with Cone Beam CT (CBCT) image data. Point matching algorithms are combined with Spline-based transforms for this purpose. The initial result of different combination is evaluated with synthetical image data.

Siming Bayer, Roman Schaffert, Nishant Ravikumar, Andreas Maier, Xiaodong Tong, Hu Wang, Martin Ostermeier, Rebecca Fahrig
Abstract: Patches in Magnetic Particle Imaging

Magnetic Particle Imaging (MPI) is a tracer-based imaging technology [1] with which superparamagnetic nanoparticles can be detected and located using specific magnetic fields. The selection field, a gradient field in form of a field free point (FFP), restricts the area in which particles can be remagnetized. The drive field, a homogeneous and time-varying magnetic field, remagnetizes the particles and moves the FFP. As a result a time dependent signal can be measured and then reconstructed to the actual spatial distribution of the tracer.

Mandy Ahlborg, Christian Kaethner, Patryk Szwargulski, Tobias Knopp, Thorsten M. Buzug
Phasenkontrast Röntgen mit 2 Phasengittern und medizinisch relevanten Detektoren

Kurzfassung. In den letzten Jahren hat die Forschung zu gitterbasierter Phasenkontrast-Bildgebung mittels Röntgenstrahlen große Fortschritte gemacht. Neueste Ergebnisse zeigen, dass das Absorptionsgitter G2, durch ein zweites phasenschiebendes Gitter ersetzt werden kann und somit die Absorption hinter dem Patienten vermieden wird. Durch die Überlagerung des Selbstabbildes des ersten Phasengitters mit dem zweiten Phasengitter wird eine Schwebung erzeugt, deren Periode ausreichend groß ist, um mit dem Detektor direkt aufgelöst zu werden. In diesem Beitrag wollen wir diesen sogenannten zwei Phasengitteraufbau analysieren. Insbesondere untersuchen wir die Möglichkeiten, solche Aufbauten mit medizinisch relevanten Detektoren zu realisieren. Einen großen Einfluss auf die Ergebnisse hat hierbei die Impulsantwort des Detektors. Mit diesem Wissen wurde die Schwebungsfrequenz bestimmt, die eine möglichst hohe Visibilität liefert. Um die simulierten Ergebnisse zu validieren, wurden Messungen analog zu den Simulationen durchgeführt. Die Ergebnisse der Simulation und der Messungen stimmen sehr gut überein.

Johannes Bopp, Michael Gallersdörfer, Veronika Ludwig, Maria Seifert, Andreas Maier, Gisela Anton, Christian Riess
CT-basiertes virtuelles Fräsen am Felsenbein
Bild- und haptischen Wiederholfrequenzen bei unterschiedlichen Rendering Methoden

Kurzfassung. Im Rahmen der Entwicklung eines haptisch-visuellen Trainingssystems für das Fräsen am Felsenbein werden ein Haptikarm und ein autostereoskopischer 3D-Monitor genutzt, um Chirurgen die virtuelle Manipulation von knöchernen Strukturen im Kontext eines sog. Serious Game zu ermöglichen. Unter anderem sollen Assistenzärzte im Rahmen ihrer Ausbildung das Fräsen am Felsenbein für das chirurgische Einsetzen eines Cochlea-Implantats üben können. Die Visualisierung des virtuellen Fräsens muss dafür in Echtzeit und möglichst realistisch modelliert, implementiert und evaluiert werden. Wir verwenden verschiedene Raycasting Methoden mit linearer und Nearest Neighbor Interpolation und vergleichen die visuelle Qualität und die Bildwiederholfrequenzen der Methoden. Alle verglichenen Verfahren sind sind echtzeitfähig, unterscheiden sich aber in ihrer visuellen Qualität.

Daniela Franz, Maria Dreher, Martin Prinzen, Matthias Teßmann, Christoph Palm, Uwe Katzky, Jerome Perret, Mathias Hofer, Thomas Wittenberg
Segmentierung von Brustvolumina in Magnetresonanztomographiedaten unter der Verwendung von Deep Learning

Kurzfassung. Die Segmentierung von Hintergrund und Brustgewebe ist ein wichtiger Teil der Auswertung von Magnetresonanztomographie-Daten der Brust. Normalerweise wird diese von Ärzten manuell durchgeführt. In dieser Arbeit wurde die Segmentierung hingegen mit einer U-net Architektur realisiert. Dabei wurden zwei Netzwerke trainiert und anschließend auf ein unbekanntes Testset, bestehend aus 8 Probandinnen, angewendet. Die so berechneten Segmentierungen wurden dann mit von Ärzten manuell vorgenommenen verglichen. Das erste U-net nutzt keine weitere Vorverarbeitungsmethode und erreicht einen DSC von 0.91±0.09 (Mittelwert ± Standardabweichung). Beim zweiten Netzwerk wurde der N4ITK Bias Correction Algorithmus als Vorverarbeitungsmethode verwendet. Die Masken für N4ITK können sehr grob sein und daher in einer späteren Anwendung von einem Arzt schnell erstellt werden. In dieser Konstellation wurde bei der Segmentierung des Testsets ein DSC von 0.98±0.05 erreicht. Die Segmentierungen benötigen darüber hinaus nach Anfertigung der Masken für den Vorverarbeitungsalgorithmus 14s. Die Methode hat somit das Potential, Anwendung in der medizinischen Diagnostik zu finden.

Tatyana Ivanovska, Thomas G. Jentschke, Katrin Hegenscheid, Henry Völzke, Florentin Wörgötter
Einfluss nicht-rigider Bildregistrierung auf 4D-Dosissimulation bei extrakranieller SBRT

Kurzfassung. Das Ziel dieser Studie ist der Vergleich nicht-rigider Open Source-Registrierungsalgorithmen (DIR) in Bezug auf ihre Genauigkeit sowie ihren Einfluss auf korrespondenzmodellbasierte 4D-Dosissimulation bei extrakranieller Strahlentherapie (SBRT). Es wurden drei verbreitete DIR-Algorithmen ausgewählt und mittels der DIRLAB-4D-CTDatensätze zunächst ihre Registrierungsgenauigkeit evaluiert sowie Korrespondenzmodelle (regressionsbasierte Korrelation von externen Atemsignalmessungen und internen Bewegungsfeldern) generiert und die Modellpräzision analysiert. Unter Verwendung von zehn Strahlentherapie-Planungs-4D-CT-Datensätzen von fünf Leber- und fünf Lungen-Tumorpatienten wurden dann Korrespondenzmodelle gebildet und im Rahmen einer modellbasierten 4D-Dosissimulation zur Abschätzung der Auswirkungen der patientenindividuellen Bewegungen während der Bestrahlung auf die applizierte Dosis eingesetzt. Berechnete Abweichungen zwischen geplanter und 4D-simulierter Dosisverteilung wurden verglichen und mit der Registrierungsgenauigkeit sowie bekannten klinischen Endpunkten (Lokalrezidiv ja/nein) in Beziehung gesetzt.

Nik Mogadas, Thilo Sothmann, René Werner
Effiziente Segmentierung trachealer Strukturen in MRI-Aufnahmen

Kurzfassung. Die Segmentierung verschiedener Strukturen im Körper ist eine der grundlegenden Operationen in der medizinischen Bildverarbeitung. In dieser Arbeit werden auf Machine Learning basierende Methoden zur Segmentierung medizinischer Bilder untersucht. Das Ziel ist es, in MRI-Scans die Trachea zu segmentieren. Jedoch soll in dieser Arbeit speziell die Effizienz der Algorithmen im Vordergrund stehen. Die verwendeten Ansätze basierten auf einer Deep Learning Architektur, welche zunächst individuell optimiert wird. Es konnte ein maximaler DICE-Koeffizient von (94.4±2.1)% erzielt werden. Zusätzlich kann festgestellt werden, dass die Segmentierung sehr effizient geschieht. Die Segmentierung von einmen Datensatz aus 40 Schichten dauert dabei weniger als eine Sekunde, wobei bei bisherigen Methoden es über eine Minute benötigte.

Tatyana Ivanovska, Philip Dietrich, Catherine Schmidt, Henry Völzke, Achim Beule, Florentin Wörgötter
Abstract: Populationsbasierte 4D Bewegungsatlanten für VR Simulationen

Atembewegte Avatare können in einem kürzlich vorgestellten visuo-haptischen Virtual Reality (VR) 4D-Simulatorkonzept modelliert [1] und GPU-basiert dargestellt [2] werden. Nadelinterventionsimulationen im hepatischen Bereich mit atmenden virtuellen Patientenkörpern sind aktuell ohne die patientenspezifische, dosisrelevante 4D-Datenerfassung nicht durchführbar. Hierbei kann ein populationsbasierter Ansatz zur Modellierung eines gemittelten, übertragbaren (4DAtembewegungsatlas) abhelfen und die Risiken einer dosisrelevanten und teuren Erfassung eines 4D-Datensatzes mindern [3].

Andre Mastmeyer, Matthias Wilms, Heinz Handels
Abstract: Rekonstruktion der initialen Druckverteilung photoakustischer Bilder mit limitiertem Blickwinkel durch maschinelle Lernverfahren

Die Rekonstruktion von Bildern aus unvollständigen Rohdaten ist eine fundamentale Herausforderung in der medizinischen Bildgebung. Dies gilt insbesondere auch für die Photoakustik, einer neuartigen Bildgebungstechnik, welche auf dem photoakustischen Effekt basiert, bei dem durch die Absorption von Photonen aus Laserpulsen im Gewebe Schallwellen ausgelöst werden. Durch den optischen Kontrast der Photoakustik können funktionale Parameter - wie die Blutsauerstoffsättigung - hoch aufgelöst und tief im Gewebe gemessen werden.

Dominik Waibel, Janek Gröhl, Fabian Isensee, Klaus Hermann Maier-Hein, Lena Maier-Hein
Abstract: Erweiterung des Bildgebungsbereiches bei der Magnetpartikelbildgebung durch externe axiale Verschiebungen

Die Magnetpartikelbildgebung (engl. Magnetic-Particle-Imaging, MPI) ist eine Bildgebungsmodalität, die auf der Darstellung super-paramagnetischer Nanopartikeln unter Verwendung von statischen und dynamischen Magnetfeldern basiert [1]. Das Bildgebungsverfahren weist eine hohe zeitliche Auflӧsung von über 40 Volumen pro Sekunde auf und ist mit einer Detektionsgrenze von 5 ng Eisen [2] ein hӧchst sensitives Verfahren, mit dem viele medizinische Applikationen adressiert werden kӧnnen. Als Einschränkung ist der, aus Sicherheitsgründen [3] auf wenige Kubikcentimeter beschränkte, Bildgebungsbereich zu nennen.

Patryk Szwargulski, Nadine Gdaniec, Matthias Graeser, Martin Mӧddel, Florian Griese, Tobias Knopp
Abstract: Random-Forest-basierte Segmentierung der subkutanen Fettschicht der Mäusehaut in 3D-OCT-Bilddaten

Die Kryolipolyse ist ein nichtinvasives kosmetisches Verfahren zur lokalen Fettreduktion [1], bei der durch kontrollierte Kühlung selektiv subkutane Fettzellen zerstӧrt werden. Für eine quantitative Evaluation des Verfahrens soll die subkutane Fettschicht in Mäusen segmentiert werden. Für eine Darstellung der Mäusehaut wurde die Optische Kohärenztomographie (OCT) als Bildmodalität genutzt, die eine detaillierte Aufnahme der subkutanen Fettschicht in Mikrometer-Auflӧsung ermӧglicht.

Timo Kepp, Christine Droigk, Malte Casper, Michael Evers, Nunciada Salma, Dieter Manstein, Heinz Handels
Towards Whole-body CT Bone Segmentation

Bone segmentation from CT images is a task that has been worked on for decades. It is an important ingredient to several diagnostics or treatment planning approaches and relevant to various diseases. As high-quality manual and semi-automatic bone segmentation is very time-consuming, a reliable and fully automatic approach would be of great interest in many scenarios. In this publication, we propose a UNet inspired architecture to address the task using Deep Learning. We evaluated the approach on whole-body CT scans of patients suffering from multiple myeloma. As the disease decomposes the bone, an accurate segmentation is of utmost importance for the evaluation of bone density, disease staging and localization of focal lesions. The method was evaluated on an in-house data-set of 6000 2D image slices taken from 15 whole-body CT scans, achieving a dice score of 0.96 and an IOU of 0.94.

André Klein, Jan Warszawski, Jens Hillengaß, Klaus Hermann Maier-Hein
Ideal Seed Point Location Approximation for GrowCut Interactive Image Segmentation

The C-arm CT X-ray acquisition process is a common modality in medical imaging. After image formation, anatomical structures can be extracted via segmentation. Interactive segmentation methods bear the advantage of a dynamically adjustable trade-off between time and achieved segmentation quality for the object of interest w.r.t. fully automated approaches. The segmentation’s quality can be measured in terms of the Dice coefficient with the ground truth segmentation image. A user’s interaction traditionally consist of drawing pictorial hints on an overlay image to the acquired image data via a graphical user interface (UI). The quality of a segmentation utilizing a set of drawn seeds varies depending on the location of the seed points in the image. In this paper, we (1) investigate the influence of seed point location on segmentation quality and (2) propose an approximation framework for ideal seed placements utilizing an extension of the well established GrowCut segmentation algorithm and (3) introduce a user interface for the utilization of the suggested seed point locations. An extensive evaluation of the predictive power of seed importance is conducted from hepatic lesion input images. As a result, our approach suggests seed points with a median of 72.5% of the ideal seed points’ associated Dice scores, which is an increase of 8.4% points to sampling the seed location at random.

Mario Amrehn, Maddalena Strumia, Stefan Steidl, Tim Horz, Markus Kowarschik, Andreas Maier
Transfer Learning for Breast Cancer Malignancy Classification based on Dynamic Contrast-Enhanced MR Images

In clinical contexts with very limited annotated data, such as breast cancer diagnosis, training state-of-the art deep neural networks is not feasible. As a solution, we transfer parameters of networks pretrained on natural RGB images to malignancy classification of breast lesions in dynamic contrast-enhanced MR images. Since DCE-MR images comprise several contrasts and timepoints, a direct finetuning of pretrained networks expecting three input channels is not possible. Based on the hypothesis that a subset of the acquired image data is sufficient for a computer-aided diagnosis, we provide an experimental comparison of all possible subsets of MR image contrasts and determine the best combination for malignancy classification. A subset of images acquired at three timepoints of dynamic T1-weighted images which closely corresponds to human interpretation performs best with an AUC of 0.839.

Christoph Haarburger, Peter Langenberg, Daniel Truhn, Hannah Schneider, Johannes Thüring, Simone Schrading, Christiane K. Kuhl, Dorit Merhof
Traditional Machine Learning Techniques for Streak Artifact Reduction in Limited Angle Tomography

In this work, the application of traditional machine learning techniques, in the form of regression models based on conventional, “hand-crafted” features, to streak reduction in limited angle tomography is investigated. Specifically, linear regression (LR), multi-layer perceptron (MLP), and reduced-error pruning tree (REPTree) are investigated. When choosing the mean-variation-median (MVM), Laplacian, and Hessian features, REPTree learns streak artifacts best and reaches the smallest root-mean-square error (RMSE) of 29HU for the Shepp-Logan phantom. Further experiments demonstrate that the MVM and Hessian features complement each other, whereas the Laplacian feature is redundant in the presence of MVM. Preliminary experiments on clinical data suggests that further investigation of clinical applications using REPTree may be worthwhile.

Yixing Huang, Yanye Lu, Oliver Taubmann, Guenter Lauritsch, Andreas Maier
Classification of Polyethylene Particles and the Local CD3+ Lymphocytosis in Histological Slices

In 2014, about 400.000 endoprosthetic operations were performed in Germany [1]. Unfortunately, the lifespan is limited and already after 10 years 5 percent of the patients have primary complaints [2]. All the more important it is to clarify the causes for this failure. One main cause is an immune response to abrasion particles of the implant, an effect which is assumed to be correlated with occurrence and count of CD3+ immune/inflammatory cells [3]. For the further analysis of this effect, computer-aided classification and image analysis methods provide a high value for the medical research. Aim of this work was the development of an threshold-based algorithm for the segmentation of polyethylene abrasion particles and the CD3+ immune/inflammatory response of histological slice images.

Lara-Maria Steffes, Marc Aubreville, Stefan Sesselmann, Veit Krenn, Andreas Maier
Synthetic Fundus Fluorescein Angiography using Deep Neural Networks

Fundus fluorescein angiography yields complementary image information when compared to conventional fundus imaging. Angiographic imaging, however, may pose risks of harm to the patient. The output from both types of imaging have different characteristics, but the most prominent features of the fundus are shared in both images. Thus, the question arises if conventional fundus images alone provide enough information to synthesize an angiographic image. Our research analyzes the capacity of deep neural networks to synthesize virtual angiographic images from their conventional fundus counterparts.

Florian Schiffers, Zekuan Yu, Steve Arguin, Andreas Maier, Qiushi Ren
Hippocampus Segmentation and SPHARM Coefficient Selection are Decisive for MCI Detection

Spherical Harmonics (SPHARM), when computed from hippocampus segmentation, have been shown to be useful features for discriminatingMCI affected patients from healthy controls. In this paper we assess the impact (i) of using different hippocampus segmentation techniques, among them three out-of-the-box automated segmentation tools and three human raters with different qualification, and (ii) of applying different strategies which SPHARM coefficients to submit to SVM-based two-class classification. We find that both choices are crucial for successful classification.

A. Uhl, M. Liedlgruber, K. Butz, Y. Höller, G. Kuchukhidze, A. Taylor, A. Thomschevski, O. Tomasi, E. Trinka
Classification of Mitotic Cells
Potentials Beyond the Limits of Small Data Sets

Tumor diagnostics are based on histopathological assessments of tissue biopsies of the suspected carcinogen region. One standard task in histopathology is counting of mitotic cells, a task that provides great potential to be improved in speed, accuracy and reproducability. The advent of deep learning methods brought a significant increase in precision of algorithmic detection methods, yet it is dependent on the availability of large amounts of data, completely capturing the natural variability in the material. Fully segmented images are provided by the MITOS dataset with 300 mitotic events. The ICPR2012 dataset provides 326 mitotic cells and in AMIDA2014 dataset, 550 mitotic cells for training and 533 for testing. In contrast to these datasets, a dataset with high number of mitotic events is missing. For this, either one of two pathologist annotated at least 10 thousand cell images for cells of the type mitosis, eosinophilic granulocyte and normal tumor cell from canine mast cell tumor whole-slide images, exceeding all publicly available data sets by approximately one order of magnitude. We tested performance using a standard CNN approach and found accuracies of up to 0.93.

Maximilian Krappmann, Marc Aubreville, Andereas Maier, Christof Bertram, Robert Klopfleisch
Markerless Coil Classification and Localization in a Routine MRI Examination Setting using an RGB-D Camera

In a routine MRI scan, a radio-frequency (RF) coil must be selected and placed around the region of interest (ROI). This is a crucial step in the workflow as the accurate coil placement is paramount for obtaining high-quality images. However, in the existing workflow, the position of the coil placement on the patient is estimated empirically by the medical technical assistant (MTA). This routine coil placement process has two shortcomings. On the one hand, the expertise of MTA in coil placement, taking the anatomical difference between patients into account, have a huge impact on the accuracy of the coil placement, and subsequently the image quality. On the other hand, the risk of selecting and placing the incorrect coil should be also be acknowledged. To improve the current workflow and provide feedback ahead of the MRI scans, we use an RGB-D camera to acquire extra information. Using the depth images taken before and after placing the coil, we propose a novel method to classify the coil type and localize the coil position during the coil placement process such that the MTA can place the coil correctly and accurately. We trained and evaluated our method over 100 synthetic data sets. We used two types of coils and placed and deformed them differently according to the anatomical region. The evaluation shows that we can classify the coil type without any error, and localize the coil with a mean translational error of 7.1 cm and mean rotation angle error of 0.025 rad.

Janani G. Nadar, Xia Zhong, Andreas Maier
Comparative Analysis of Unsupervised Algorithms for Breast MRI Lesion Segmentation

Accurate segmentation of breast lesions is a crucial step in evaluating the characteristics of tumors. However, this is a challenging task, since breast lesions have sophisticated shape, topological structure, and variation in the intensity distribution. In this paper, we evaluated the performance of three unsupervised algorithms for the task of breast Magnetic Resonance (MRI) lesion segmentation, namely, Gaussian Mixture Model clustering, K-means clustering and a markercontrolled Watershed transformation based method. All methods were applied on breast MRI slices following selection of regions of interest (ROIs) by an expert radiologist and evaluated on 106 subjects’ images, which include 59 malignant and 47 benign lesions. Segmentation accuracy was evaluated by comparing our results with ground truth masks, using the Dice similarity coefficient (DSC), Jaccard index (JI), Hausdorff distance and precision-recall metrics. The results indicate that the marker-controlled Watershed transformation outperformed all other algorithms investigated.

Sulaiman Vesal, Nishant Ravikumar, Stephan Ellman, Andreas Maier
Segmentation of Fat and Fascias in Canine Ultrasound Images

The connective tissue between fat and muscle termed fascia has been of interest to the recent clinical and biological research. However, in the canine and human medicine, the anatomic knowledge is still limited. To analyze the superficial fascia in canine medicine, a database with around 200 ultrasound images of one dog has been created. The superficial fascia contains fat compartments and is closely connected to the surrounding structures such as the skin’s dermis and the epimysium of the muscles. This work proposes a semi-automatic and fully-automatic segmentation algorithm separating the different layers of ultrasound images of canine. Both algorithms were evaluated on a set of 24 expert-labeled images achieving high accuracy scores up to 95.9%.

Oleksiy Rybakov, Daniel Stromer, Irina Mischewski, Andreas Maier
Manifold Learning-based Data Sampling for Model Training

Training data sampling is an important task in machine learning especially for data with small sample size and data with nonuniform sample distribution. Dividing data into different data sets randomly can cause the problem that, the training model covers only parts of the sampled cases and works inaccurately for weakly sampled cases. Recent research showed the benefit of manifold learning techniques in medical image processing. In this work, we propose a manifold learning based approach to improve the data division and the model training. We evaluated the proposed approach using an atlas registration framework and a deep learning framework. The final segmentation results using methods with and without data balancing were compared. All of the final segmentations were improved after implementing the manifold learning based approach into the frameworks. The largest improvement was 24.4%. Thus, the proposed manifold learning based approach is effective for the model training.

Shuqing Chen, Sabrina Dorn, Michael Lell, Marc Kachelrieß, Andreas Maier
Computer-aided Detection of the Most Suitable MRI Sequences for Subsequent Spinal Metastasis Delineation

Detection and segmentation of vertebral metastases is a crucial step for support of diagnosis and treatment planning, especially in minimally invasive interventions. Even though computer-assistant tools will not dispense radiologists yet, algorithmically supported detection and segmentation of spinal metastases will play a more and more important role in the near future. The usage of images, where a sufficiently good differentiation between metastases and surrounding tissue is possible, constitutes a critical requirement for successful segmentation procedures. Therefore, we proposed a pipeline, that semi-automatically sorts out unsuitable imaging sequences, as well as combinations of different images via absolute intensity difference images and returns a ranking based recommendation of which image data fits best the requirements for future segmentation tasks. We evaluated our method with 10 patient cases and matched the produced ranking with those of a segmentation field expert. With an average Spearman’s ranking coefficient of 0.92±0.07, our method showed promising results and could be a valueable pre-processing step to speed up clinical segmentation procedures due to omitting the time-consuming manual initialization of choosing suitable image data.

Georg Hille, Steffen Serowy, Klaus Tönnies, Sylvia Saalfeld
Abstract: Deep Residual Learning for Limited Angle Artefact Correction

Using non-conventional scan trajectories for Cone Beam (CB) imaging promise low dose interventions and radiation protection to the personal [1]. The here investigated circular tomosynthesis yields good image quality in two preferred directions, but introduces limited angle artefacts in the third. The artefacts become more severe, the smaller the half tomo angle α gets.

Alena-K. Schnurr, Khanlian Chung, Lothar R. Schad, Frank G. Zöllner
Abstract: AngioUnet
A Convolutional Neural Network for Vessel Segmentation in Cerebral DSA Series

Das U-net [1] ist eine vielversprechende Architektur für Segmentierungsprobleme im Bereich der Medizin. Wir zeigen, wie sich diese Architektur effektiv auf die Segmentierung von zerebralen DSA Zeitserien anwenden lässt. Durch die Erweiterung der Eingabe auf mehrere Bilder wird es möglich besser zwischen Gefäsen und Hintergrund zu unterscheiden.

Christian Neumann, Klaus-D. Tönnies, Regina Pohle-Fröhlich
Abstract: Measuring Muscle Contractions from Single Element Transducer Ultrasound Data using Machine Learning Strategies

Being aware of the correct execution of certain fitness exercises (e.g. squats) is important for rehabilitation and sports athletes alike. Thus, being able to noninvasively distinguish between contracted and relaxed muscle states is crucial. Measurements using optical systems (movement tracking) or kinetic approaches (muscle circumference) only provide information from the body surface with varying accuracy.

Lukas Brausch, Holger Hewener
Abstract: Automated Segmentation of Bones for the Age Assessment in 3D MR Images using Convolutional Neural Networks

The age assessment is a complicated procedure used to determine the chronological age of an individual who lacks legal documentation. Actual studies show that the ossification degree of the growth plates in the knee joint represents a suitable indicator for the majority age. To verify this hypothesis a high number of datasets have to be analysed.

Markus Auf-der-Mauer, Paul-Louis Pröve, Eilin Jopp, Jochen Herrmann, Michael Groth, Michael M. Morlock, Ben Stanczus, Dennis Säring
Abstract: OCT-OCTA Segmentation
A Novel Framework and an Application to Segment Bruch’s Membrane in the Presence of Drusen

In this work, a novel paradigm for segmenting optical coherence tomography (OCT) and optical coherence tomography angiography (OCTA) is presented [1]. Since it uses OCT and OCTA information jointly it is called “OCT-OCTA segmentation” and its usefulness is demonstrated by segmenting the Bruch’s Membrane (BM) in the presence of drusen. Therefore a fully automatic graph-cut algorithm was developed and evaluated by comparing the automatic segmentation results with manual segmentation in 7 eyes (6 patients; 73.8 ± 5.7 y/o) with nascent geographic atrophy and/or drusen associated geographic atrophy.

Julia Schottenhamml, Eric M. Moult, Eduardo A. Novais, Martin F. Kraus, ByungKun Lee, WooJhon Choi, Stefan B. Ploner, Lennart Husvogt, Chen D. Lu, Patrick Yiu, Philip J. Rosenfeld, Jay S. Duker, Andreas K. Maier, Nadia Waheed, James G. Fujimoto
Human Pose Estimation from Pressure Sensor Data

In-bed motion monitoring has become of great interest for a variety of clinical applications. In this paper, we introduce a hashbased learning method to retrieve human poses from pressure sensors data in real time considering temporal correlation between poses. The basis of our approach is a multimodal database describing different in-bed activities. Database entries have been created using an array of pressure sensors and an additional motion capture system. Our results show good performance even in poses where the subject has minimal contact with the sensors

Leslie Casas, Chris Mürwald, Felix Achilles, Diana Mateus, Dietrich Huber, Nassir Navab, Stefanie Demirci
Force-feedback-assisted Bone Drilling Simulation Based on CT Data

In order to fix a fracture using minimally invasive surgery approaches, surgeons are drilling complex and tiny bones with a 2 dimensional X-ray as single imaging modality in the operating room. Our novel haptic force-feedback and visual assisted training system will potentially help hand surgeons to learn the drilling procedure in a realistic visual environment. Within the simulation, the collision detection as well as the interaction between virtual drill, bone voxels and surfaces are important. In this work, the chai3d collision detection and force calculation algorithms are combined with a physics engine to simulate the bone drilling process. The chosen Bullet-Physics-Engine provides a stable simulation of rigid bodies, if the collision model of the drill and the tool holder is generated as a compound shape. Three haptic points are added to the K-wire tip for removing single voxels from the bone. For the drilling process three modes are proposed to emulate the different phases of drilling in restricting the movement of a haptic device.

Johannes Maier, Michaela Huber, Uwe Katzky, Jerome Perret, Thomas Wittenberg, Christoph Palm
Direct Volume Rendering in Virtual Reality

Direct Volume Rendering (DVR) techniques are used to visualize surfaces from 3D volume data sets, without computing a 3D geometry. Several surfaces can be classified using a transfer function by assigning optical properties like color and opacity (RGBα) to the voxel data. Finding a good transfer function in order to separate specific structures from the volume data set, is in general a manual and time-consuming procedure, and requires detailed knowledge of the data and the image acquisition technique. In this paper, we present a new Virtual Reality (VR) application based on the HTC Vive headset. Onedimensional transfer functions can be designed in VR while continuously rendering the stereoscopic image pair through massively parallel GPUbased ray casting shader techniques. The usability of the VR application is evaluated.

Ingrid Scholl, Sebastian Suder, Stefan Schiffer
Automatic Multi-modal Cervical Spine Image Atlas Segmentation
Using Adaptive Stochastic Gradient Descent

A personalized medicine has been advanced in different fields of medicine to combat, control and prevent a number of diseases. In personalized medicine, products are customized and only suitable for a specific patient. In spinal surgery, medical images are taken into account to implant spinal devices with the aim of minimizing the risk of insufficient implant fit. A model of the spine is generated from these images and used in biomechanic framework to simulate the effect of the customized implant on a specific patient.To generate such a model, an efficient and practical segmentation method is needed which is proposed in this paper. The large deformation of human spine and the touching boundaries of neighboring vertebrae make the problem of spine segmentation very challenging. The classical segmentation methods e.g. thresholding or region growing fail to separate different vertebrae. The state of the art methods using shape models require a long time for training and testing. A new method for automatic multi-modal cervical spine segmentation is proposed in this paper. The proposed method requires only a few seconds to segment a specific vertebra or the whole cervical spine. It is provided as Slicer 3D plug-in which is free and open-source. The public datasets available fail to provide high quality MRI cervical spine images. Another contribution of this study is providing a high quality multi-modal cervical spine public and free dataset.

Ibraheem Al-Dhamari, Sabine Bauer, Dietrich Paulus
SlideRunner
A Tool for Massive Cell Annotations in Whole Slide Images

Large-scale image data such as digital whole-slide histology images pose a challenging task at annotation software solutions. Today, a number of good solutions with varying scopes exist. For cell annotation, however, we find that many do not match the prerequisites for fast annotations. Especially in the field of mitosis detection, it is assumed that detection accuracy could significantly benefit from larger annotation databases that are currently however very troublesome to produce. Further, multiple independent (blind) expert labels are a big asset for such databases, yet there is currently no tool for this kind of annotation available. To ease this tedious process of expert annotation and grading, we introduce SlideRunner, an open source annotation and visualization tool for digital histopathology, developed in close cooperation with two pathologists. SlideRunner is capable of setting annotations like object centers (for e.g. cells) as well as object boundaries (e.g. for tumor outlines). It provides single-click annotations as well as a blind mode for multi-annotations, where the expert is directly shown the microscopy image containing the cells that he has not yet rated.

Marc Aubreville, Christof Bertram, Robert Klopfleisch, Andreas Maier
Automated Containerized Medical Image Processing Based on MITK and Python
A Modular System to Implement Medical Image Processing Pipelines and Visualize Meta Data

Modern medical image processing employs an ever widening array of tools on an ever larger pool of data. Development of the tools takes place on a variety of different platforms, determined by external circumstances, such as the availability of necessary functionality as libraries forcing the use of a specific programming language or interdependencies of third party tools requiring specific versions of specific operating systems. This problem itself is not unique to medical imaging[1] and solutions such as Docker have proven themselves useful in a variety of use cases.

Caspar J. Goch, Jasmin Metzger, Martin Hettich, André Klein, Tobias Norajitra, Michael Götz, Jens Petersen, Klaus H. Maier-Hein, Marco Nolden
Multi-channel Deep Transfer Learning for Nuclei Segmentation in Glioblastoma Cell Tissue Images

Segmentation and quantification of cell nuclei is an important task in tissue microscopy image analysis. We introduce a deep learning method leveraging atrous spatial pyramid pooling for cell segmentation. We also present two different approaches for transfer learning using datasets with a different number of channels. A quantitative comparison with previous methods was performed on challenging glioblastoma cell tissue images. We found that our transfer learning method improves the segmentation result.

Thomas Wollmann, Julia Ivanova, Manuel Gunkel, Inn Chung, Holger Erfle, Karsten Rippe, Karl Rohr
Stitching Pathological Tissue Images using DOP Feature Tracking

This contribution introduces an approach for stitching multiple images of a histological slide to a panorama image using Differences of Paraboloids (DOP). DOP provides a novel method for the detection, description and matching of features of two overlapping images. In our context of manual whole-slide imaging (WSI), DOP extracts essential keypoints of an image and describes them with feature vectors considering the keypoint’s neighborhood. The DOP feature vector of the current image is then matched against the feature vectors of all previous images. With matching correspondences, a feature based image registration is generated that estimates the translation between two overlapping images. Likewise, all images are aligned to form a whole-slide panorama. Our results reveal a superior stitching quality employing the presented DOP approach in comparison to the well-known SIFT and SURF. Our evaluation is based on the homogeneity at the artifically created edges in the panorama due to the stitching. The DOP offers a convincing solution to stitch pathological tissue.

Matthias Bergler, Maximilian Weiherer, Tobias Bergen, Malte Avenhaus, David Rauber, Thomas Wittenberg, Christian Münzenmayer, Michaela Benz
Motion Artifact Detection in Confocal Laser Endomicroscopy Images

Confocal Laser Endomicroscopy (CLE), an optical imaging technique allowing non-invasive examination of the mucosa on a (sub)- cellular level, has proven to be a valuable diagnostic tool in gastroenterology and shows promising results in various anatomical regions including the oral cavity. Recently, the feasibility of automatic carcinoma detection for CLE images of sufficient quality was shown. However, in real world data sets a high amount of CLE images is corrupted by artifacts. Amongst the most prevalent artifact types are motion-induced image deteriorations. In the scope of this work, algorithmic approaches for the automatic detection of motion artifact-tainted image regions were developed. Hence, this work provides an important step towards clinical applicability of automatic carcinoma detection. Both, conventional machine learning and novel, deep learning-based approaches were assessed. The deep learning-based approach outperforms the conventional approaches, attaining an AUC of 0.90.

Maike Stoeve, Marc Aubreville, Nicolai Oetter, Christian Knipfer, Helmut Neumann, Florian Stelzle, Andreas Maier
Variational Networks for Joint Image Reconstruction and Classification of Tumor Immune Cell Interactions in Melanoma Tissue Sections

Immunotherapy is currently revolutionizing the treatment of cancer. Detailed analyses of tumor immune cell interaction in the tumor microenvironment will facilitate an accurate prediction of a patient’s clinical response. The automatic and reliable pre-screening of histological tissue sections for tumor infiltrating immune cells (TILs) will support the development of TIL-based predictive biomarkers for checkpoint immunotherapy. In this paper, a learning approach for image classification is presented, which allows various pattern inquires for different types of tissue section images. The underlying trainable reaction diffusion model combines classification and denoising. The model is trained using a stochastic generation of training data. The effectiveness of this approach is demonstrated for immunofluorescent and for Hematoxylin and Eosin (H&E) stained melanoma section images. A particular focus is on the classification of TILs in the proximity to melanoma cells in an experimental melanoma mouse model and in human melanoma. This new learning approach for images of melanoma tissue sections will refine the strategy for the practical clinical application of biomarker research.

Alexander Effland, Michael Hölzel, Teresa Klatzer, Erich Kobler, Jennifer Landsberg, Leonie Neuhäuser, Thomas Pock, Martin Rumpf
Frangi-Net

In this paper, we reformulate the conventional 2-D Frangi vesselness measure into a pre-weighted neural network (“Frangi-Net”), and illustrate that the Frangi-Net is equivalent to the original Frangi filter. Furthermore, we show that, as a neural network, Frangi-Net is trainable. We evaluate the proposed method on a set of 45 high resolution fundus images. After fine-tuning, we observe both qualitative and quantitative improvements in the segmentation quality compared to the original Frangi measure, with an increase up to 17% in F1 score.

Weilin Fu, Katharina Breininger, Roman Schaffert, Nishant Ravikumar, Tobias Würfl, Jim Fujimoto, Eric Moult, Andreas Maier
Lung Vessel Enhancement in Low-Dose CT Scans
The LANCELOT Method

To reduce the patient’s radiation exposure from computed tomography scans (CT), low-dose CT scans can be recorded. Several image processing methods exist to segment or enhance the lung blood vessels from contrast-enhanced or high resolution CT scans, but the reduced contrast in low-dose CT scans leads to over- or under-segmentation. Our LANCELOT method combines maximum response and stick filters to enhance lung blood vessels in native, low-dose CT scans. We compare our method with the vessel segmentation and enhancing methods from Frangi and Sato et al. Our method has two advantages that were confirmed in an evaluation with two clinical experts: First, our method enhances small vessels and vessel branches more clearly and second, it connects vessels anatomically correct, while the others create discontinuities.

Nico Merten, Kai Lawonn, Philipp Gensecke, Oliver Großer, Bernhard Preim
Whole Heart and Great Vessel Segmentation with Context-aware of Generative Adversarial Networks

Automatic segmentation of cardiac magnetic resonance imaging (CMRI) is an important application in clinical tasks. However, semantic segmentation of the myocardium and blood pool in CMR is a challenge due to differentiating branchy structures and slicing fuzzy boundaries. In this paper, we propose an automatic deep architecture for simultaneous myocardium and blood pool segmentation on patients with congenital heart disease (CHD). Inspired by vanilla generative adversarial networks (GANs), we propose a cascade of conditional GANs for semantic segmentation. The proposed cascade has three stages that are designed to share convolutional features and weights. Each stage has a conditional generative adversarial network with a unique loss function and trains on different images from the same patients. We further apply AutoContext Model to implement a context-aware generative adversarial network. The proposed method evaluated on the HVSMR dataset and the experimental results demonstrated the superior performance of our approach.

Mina Rezaei, Haojin Yang, Christoph Meinel
Impact of Gradual Vascular Deformations on the Intra-aneurysmal Hemodynamics

The treatment of intracranial aneurysms based on stentassisted coiling often leads to local vascular deformations. Patient-specific data of an aneurysm in the pre interventional and follow-up state is used to interpolate intermediate vessel-aneurysm configurations. Computational Fluid Dynamics simulations are performed in order to quantify the effect of vessel deformation on the blood flow. Results reveal gradual changes in the blood flow patterns shifting the load on the aneurysm wall from the dome to the neck region. Based on this novel concept, it is possible to virtually evaluate how different types of stents can improve or impair the treatment goal of reducing the intra-aneurysmal blood flow.

Samuel Voß, Patrick Saalfeld, Sylvia Saalfeld, Oliver Beuing, Gabor Janiga, Bernhard Preim
Myocardial Twist from X-ray Angiography
Can we Observe Left Ventricular Twist in Rotational Coronary Angiography?

We present preliminary evidence that left ventricular twist can be observed and thus estimated from rotational coronary angiography. Our method is based on an ellipsoidal parametric model initially developed for functional analysis of cardiac tagged MRI. First, we fit the model to 3D coronary artery centerlines reconstructed from rotational angiography and then use 3D/2D registration to optimize for the functional parameters driving the model. On two clinical acquisitions, we show that our method is able to recover cardiac motion indicated by an average reduction in reprojection error of 28.1±3.0%. Analysis of the functional progression of the functional parameters over time reveals radial and longitudinal contraction, and left ventricular twist. We believe that these results are exciting and encourage improvement of the proposed method in future work.

Tobias Geimer, Mathias Unberath, Johannes Höhn, Stephan Achenbach, Andreas Maier
Abstract: Retrieval of Attenuation Values by the Augmented Likelihood Image Reconstruction in the Presence of Metal Artefacts

Metal implants are able to cause severe artefacts in CT images due to physical effects such as scattering, total absorption, noise, or beamharding. Typically, the reconstructed images feature dark shadows around high-density objects as well as bright and dark streaks that may reduce the diagnostic value drastically. Within an extensive evaluation, the novel algorithm Augmented Likelihood Image Reconstruction has proven to reduce occurring artefacts accurately [1].

Maik Stille, Christian Ziemann, Florian Cremers, Dirk Rades, Thorsten M. Buzug
Abstract: Efficient Epipolar Consistency

Epipolar consistency (EC) is one of the simplest consistency conditions in cone-beam computed tomgraphy. It describes redundant line integrals between any two projection images. Its simplicity is an advantage for practical implementation and applications for calibration and motion correction in FDCT.

André Aichert, Katharina Breininger, Thomas Köhler, Andreas Maier
Magnetic-Particle-Imaging mit mehreren Gradientenstärken

Die Magnetpartikelbildgebung (engl.Magnetic-Particle-Imaging,MPI) ist ein tomografisches Bildgebungsverfahren mit dem super-paramagnetische Nanopartikel mit einer hohen zeitlichen Auflösung visualisiert werden können [1]. Die räumliche Auflösung und die Größe des Bildgebungsbereiches hängen direkt mit der genutzten Gradientenfeldstärke zusammen. Bei einer hohen Gradientenfeldstärke kann zwar eine sehr gute räumliche Auflösung erreicht werden, gleichzeitig verringert sich allerdings der Bildgebungsbereich.

Patryk Szwargulski, Nadine Gdaniec, Tobias Knopp
3D Adaptive Wavelet Shrinkage Denoising while Preserving Fine Structures

By this contribution we tackle the challenge of denoising MRI image data while preserving fine structures. Log-Gabor wavelets offer a good compromise between spatial and spectral resolution, thus allowing to extract the local phase at all image voxels. Shrinking only the complex-valued wavelet response vectors at different scales and orientations leaves the essential phase information undistorted. We propose an adaptive shrinking threshold based on supervoxels and also based on the amount of texture in a particular image region. Therefore, the proposed adaptive 3D technique preserves fine structures and outperforms existing methods in terms of peak signal-to-noise ratio (PSNR)/structural similarity (SSIM) in cases of elevated noise perturbation.

Cosmin Adrian Morariu, Alice Eckhardt, Tobias Terheiden, Stefan Landgräber, Marcus Jäger, Josef Pauli
Abstract: QuaSI – Quantile Sparse Image
A Prior for Spatio-Temporal Denoising of Retinal OCT Data

Optical Coherence Tomography (OCT) is a standard non-invasive imaging modality widely used in opthalmology. Due to its high spatial resolution OCT has become a standard imaging technique. However, speckle noise caused by photon interference during the acquisition is its major drawback.

Franziska Schirrmacher, Thomas Köhler, Lennart Husvogt, James G. Fujimoto, Joachim Hornegger, Andreas K. Maier
Erratum zu: Bildverarbeitung für die Medizin 2018

A. Maier et al. (Hrsg.), Bildverarbeitung für die Medizin 2018, Informatik aktuell, https://doi.org/10.1007/978-3-662-56537-7Die Reihenfolge der Autoren war in der Orginalversion dieser Beiträge vertauscht. Die korrekte Reihenfolge wird hier dargestellt:An Open Source Tool for Creating Model Files for Virtual Volume Rendering in PDF DocumentsJulian Brandner, Axel Newe, Wolfgang Aichinger, Linda BeckerSegmentierung von Brustvolumina in Magnetresonanztomographiedaten unter der Verwendung von Deep LearningThomas G. Jentschke, Katrin Hegenscheid, Henry Völzke, Florentin Wörgötter, Tatyana IvanovskaEffiziente Segmentierung trachealer Strukturen in MRI-AufnahmenPhilip Dietrich, Catherine Schmidt, Henry Völzke, Achim Beule, Florentin Wörgötter, Tatyana IvanovskaDer Originalbeitrag wurde korrigiert.

Andreas Maier, Thomas M. Deserno, Heinz Handels, Klaus H. Maier-Hein, Christoph Palm, Thomas Tolxdorff
Erratum zu: Amplitude of brain signals classify hunger status based on machine learning in resting-state fMRI

Erratum zu:Kapitel 13 in: A. Maier et al. (Hrsg.),Bildverarbeitung für die Medizin 2018,Informatik aktuell,https://doi.org/10.1007/978-3-662-56537-7_13An incorrect version of the article by Al-Zubaidi et al. was initially published. The original version has been retracted, and the correct version of the article has been published (under doi https://doi.org/10.1007/978-3-662-56537-7_13) online and in the print version of this book. The original version has been updated to indicate the retraction, and the correct version is now the version of record.

Arkan Al-Zubaidi, Alfred Mertins, Marcus Heldmann, Kamila Jauch-Chara, Thomas F. Münte
Backmatter
Metadata
Title
Bildverarbeitung für die Medizin 2018
Editors
Prof. Dr. Andreas Maier
Prof. Dr. Thomas M. Deserno
Prof. Dr. Heinz Handels
Dr. Klaus Hermann Maier-Hein
Prof. Dr. Christoph Palm
Prof. Dr. Thomas Tolxdorff
Copyright Year
2018
Publisher
Springer Berlin Heidelberg
Electronic ISBN
978-3-662-56537-7
Print ISBN
978-3-662-56536-0
DOI
https://doi.org/10.1007/978-3-662-56537-7

Premium Partner