Medical Image Computing and Computer-Assisted Intervention – MICCAI 2005

In this paper, we present a sensor guided ablation procedure of highly motile left atrium. It uses a system which automatically registers the 4D heart model with the position sensor on the catheter, and visualizes the heart model and the position of the catheter together in real time. With this system clinicians can easily map the motile left atrium shape and see where the catheter is inside it, therefore greatly improve the efficiency of the ablation operation.

Computer assistance for breast conserving surgery requires a guidance method to assist a surgeon in locating tumor margin accurately. A wide array of guidance methods can be considered ranging from various pictorial representations, symbolic graphical interfaces as well as those based on other sensory cues such as sound. In this study, we present an experimental framework for testing candidate guidance methods in isolation or in combination. A total of 22 guidance approaches, based on stereographic, non-stereographic, symbolic and auditory cues were tested in a simulation of breast conserving surgery. Observers were asked to circumscribe a virtual tumor with a magnetically tracked scalpel while measuring the spatial accuracy, time and the frequency with which the tumor margin was intersected. A total of 110 studies were performed with 5 volunteers. Based on these findings, we demonstrated that a single view of the tumor with a stereo presentation in conjunction with an auditory guidance cue provided the best balance of accuracy, speed and surgical integrity. This study demonstrates a practical and helpful framework for testing guidance methods in a context dependent manner.

This paper describes a system for dynamic intraoperative prostate brachytherapy using 3D ultrasound guidance with robot assistance

.

The system consists of 3D transrectal ultrasound (TRUS) imaging, a robot and software for prostate segmentation, 3D dose planning, oblique needle segmentation and tracking, seed segmentation, and dynamic re-planning and verification. The needle targeting accuracy of the system was 0.79 mm ± 0.32 mm in a phantom study.

Depth estimation is one of the most fundamental challenges for performing minimally invasive surgical (MIS) procedures. The requirement of accurate 3D instrument navigation using limited visual depth cues makes such tasks even more difficult. With the constant expectation of improving safety for MIS, there is a growing requirement for overcoming such constraints during MIS. We present in this paper a method of improving the surgeon’s perception of depth by introducing an “invisible shadow” in the operative field cast by an endoscopic instrument. Although, the shadow is invisible to human perception, it can be digitally detected, enhanced and re-displayed. Initial results from our study suggest that this method improves depth perception especially when the endoscopic instrument is in close proximity to the surface. Experiment results have shown that the method could potentially be used as an instrument navigation aid allowing accurate maneuvering of the instruments whilst minimizing tissue trauma.

The purpose of our project was to develop a novel navigation system for interventional radiology. Fields of application are minimally invasive percutaneous interventions performed under local anaesthesia. In order to reduce unintentional patient movements we used a patient vacuum immobilization device. Together with the vacuum fixation and a newly developed reference frame we achieved a fully automatic patient-to-image registration independent from the tracking system. The combination of the software and a novel designed needle holder allows for an adjustment of the needle within a few seconds. The complete system is adapted to the requirements of the radiologist and to the clinical workflow. For evaluation of the navigation system we performed a phantom study with a perspex phantom and achieved an average needle positioning accuracy of less than 0.7 mm.

A family of passive markers is presented by which the position and orientation of a surgical instrument can be computed from its ultrasound image using simple image processing. These markers address the problem of imaging instruments and tissue simultaneously in ultrasound-guided interventions. Marker-based estimates of instrument location can be used in augmented reality displays or for image-based servoing. Marker design, measurement techniques and error analysis are presented. Experimentally determined in-vitro measurement errors of 0.22 mm in position and 0.089 rad in orientation were obtained using a standard ultrasound imaging system.

Percutaneous radiofrequency ablation has become a frequently used technique for the treatment of liver cancers, but still remains very difficult to plan. In this paper, we propose a robust method to delineate on the skin of a 3D reconstructed patient the zones that are candidate for an insertion, because they allow a safe access to the tumor without meeting any organ, and to compute automatically within these zones an optimal trajectory minimizing the volume of necrosis covering the tumor.

The paper is concerned with determining the feasibility of performing telesurgery over long communication links. It describes an experimental testbed for telesurgery that is currently available in our laboratory. The tesbed is capable of supporting both wired and satellite connections as well as simulated network environments. The feasibility of performing telesurgery over a satellite link with approximately 600

ms

delay is shown through a number of dry and wet lab experiments. Quantative results of these experiments are also discussed.

This paper introduces a novel method for ultrasound calibration for both spatial and temporal parameters. The main advantage of this method is that it does not require a phantom, which is usually expensive to fabricate. Furthermore, the method does not require extensive image processing. For spatial calibration, we solve an optimization problem established by a set of equations that relate the orientations of a line (i.e., calibration pointer) to the intersection points appearing in the ultrasound image. The line orientation is provided through calibration of both ends of the calibration pointer. Temporal calibration is achieved by processing of the captured pointer orientations and the corresponding image positions of intersection along with the timing information. The effectiveness of the unified method for both spatial and temporal calibration is apparent from the quality of the 3D reconstructions of a known object.

Developing electromagnetically (EM) tracked tools can be very time consuming. Tool design traditionally takes many iterations, each of which requires construction of a physical tool and performing lengthy experiments. We propose a simulator that allows tools to be virtually designed and tested before ever being physically built. Both tool rigid body (RB) configurations and reference RB configurations are configured; the reference RB can be located anywhere in the field, and the tool is virtually moved around the reference in user-specified pattern. Sensor measurements of both RBs are artificially distorted according to a previously acquired error field mapping, and the 6-DOF frames of the Tool and Reference are refit to the distorted sensors. It is possible to predict the tool tip registration error for a particular tool and coordinate reference frame (CRF) in a particular scenario before ever even building the tools.

This paper reports evaluation of compact forceps manipulator designed for assisting laparoscopic surgery. The manipulator consists of two miniaturized parts; friction wheel mechanism which rotates and translates forceps (62×52×150[mm

3

], 0.6[kg]), and gimbals mechanism which provides pivoting motion of forceps around incision hole on the abdomen (135×165×300[mm

3

], 1.1[kg]). The four-DOF motion of forceps around the incision hole on the abdomen in laparoscopic surgery is realized. By integration with robotized forceps or a needle insertion robot, it will work as a compact robotic arm in a master-slave system. It can also work under numerical control based on the computerized surgical planning. This table-mounted miniaturized manipulator contributes to the coexistence of clinical staffs and manipulators in the today’s crowded operating room. As the results of mechanical performance evaluation with load of 4 [N], positioning accuracy was less than 1.2 [deg] in pivoting motion, less than 4 [deg] in rotation of forceps, less than 1.2 [mm] in longitudinal translation of forceps. As future works, we will modify mechanism for sterilization and safety improvement, and also integrate this manipulator with robotized forceps to build a surgery assisting robotic system.

We address the problem of the stitching task in endoscopic surgery using a circular needle under robotic assistance. Our main focus is to present an algorithm for suturing using guidance virtual fixtures (VF) that assist the surgeon to move towards a desired goal. A weighted multi-objective, constraint optimization framework is used to compute the joint motions required for the tasks. We show that with the help of VF, suturing can be performed at awkward angles without multiple trials, thus avoiding damage to tissue. In this preliminary study we show the feasibility of our approach and demonstrate the promise of cooperative assistance in complex tasks such as suturing.

For delicate operations conducted using surgical robot systems, surgeons need to receive information regarding the contact forces on the tips of surgical instruments. For the detection of this contact force, one of the authors previously proposed a new method, called the overcoat method, in which the instrument is supported by sensors positioned on the overcoat pipe. This method requires cancellation of the acceleration forces of the instrument/holder attached to the overcoat sensor. In the present report, the authors attempt to use acceleration sensors to obtain the acceleration forces of the instrument/holder. The new cancellation method provides a force-detection accuracy of approximately 0.05-0.1 N for a dynamic response range of up to approximately 20 Hz, compared to approximately 1 Hz, which was achieved by using acceleration forces based on the theoretical robot motion.

Percutaneous Vertebroplasty (PVP) is an effective and less invasive medical treatment for vertebral osteoporotic compression fractures. However, this operative procedure is quite difficult because an arcus vertebra, which is narrow, is needled with accuracy, and an operator’s hand is exposed to X-ray continuously. We have developed a needle insertion robot for Percutaneous Vertebroplasty. Its experimental evaluation on the basic performance of the system and needle insertion accuracy are presented. A needle insertion robot is developed for PVP. This robot can puncture with accuracy and an operator does not need to be exposed to X-ray. The mechanism of the robot is compact in size (350 mm × D 400 mm × H270 mm, weight: 15 kg) so that the robot system can be inserted in the space between C-arm and the patient on the operating table. The robot system is controlled by the surgical navigation system where the appropriate needle trajectory is planned based on pre-operative three-dimensional CT images. The needle holding part of the robot is X-ray lucent so that the needle insertion process can be monitored by fluoroscopy. The position of the needle during insertion process can be continuously monitored. In vitro evaluation of the system showed that average position and orientation errors were less than 1.0 mm and 1.0 degree respectively. Experimental results showed that the safety mechanism called mechanical fuse released the needle holding disk properly when excessive force was applied to the needle. These experimental results demonstrated that the developed system has the satisfactory basic performance as needle insertion robot for PVP.

For robotic assisted minimal access surgery, recovering 3D soft tissue deformation is important for intra-operative surgical guidance, motion compensation, and prescribing active constraints. We propose in this paper a method for determining varying focal lengths of stereo laparoscope cameras during robotic surgery. Laparoscopic images typically feature dynamic scenes of soft-tissue deformation and self-calibration is difficult with existing approaches due to the lack of rigid temporal constraints. The proposed method is based on the direct derivation of the focal lengths from the fundamental matrix of the stereo cameras with known extrinsic parameters. This solves a restricted self-calibration problem, and the introduction of the additional constraints improves the inherent accuracy of the algorithm. The practical value of the method is demonstrated with analysis of results from both synthetic and

in vivo

data sets.

This paper presents a hand-eye robotic model for total knee replacement (TKR) surgery. Unlike existent robot assisted TKR surgery, the proposed model is a surgical robot that combines with a movable hand-eye navigation system, which would use the full potential of both computer-assisted systems. Without using CT images and landmark pins in the patient’s bones, it can directly measure the mechanical axis with high precision. This system provides a new approach of the minimally invasive surgery. Experiment results show that the proposed model is promising in the future application.

This paper present a novel image-guided system for precise automatic targeting in keyhole minimally invasive neurosurgery. The system consists of a miniature robot fitted with a mechanical guide for needle/probe insertion. Intraoperatively, the robot is directly affixed to a head clamp or to the patient skull. It automatically positions itself with respect to predefined targets in a preoperative CT/MRI image following an anatomical registration with a intraoperative 3D surface scan of the patient facial features. We describe the preoperative planning and registration modules, and an in-vitro registration experiment of the entire system which yields a target registration error of 1.7mm (std=0.7mm).

In robotically assisted laparoscopic surgery, soft-tissue motion tracking and structure recovery are important for intraoperative surgical guidance, motion compensation and delivering active constraints. In this paper, we present a novel method for feature based motion tracking of deformable soft-tissue surfaces in totally endoscopic coronary artery bypass graft (TECAB) surgery. We combine two feature detectors to recover distinct regions on the epicardial surface for which the sparse 3D surface geometry may be computed using a pre-calibrated stereo laparoscope. The movement of the 3D points is then tracked in the stereo images with stereo-temporal constrains by using an iterative registration algorithm. The practical value of the technique is demonstrated on both a deformable phantom model with tomographically derived surface geometry and

in vivo

robotic assisted minimally invasive surgery (MIS) image sequences.

This paper presents a new algorithm for non-rigid registration between two doubly-connected regions. Our algorithm is based on harmonic analysis and the theory of optimal mass transport. It assumes an underlining continuum model, in which the total amount of mass is exactly preserved during the transformation of tissues. We use a finite element approach to numerically implement the algorithm.

In this paper we propose a new two step method to register the liver from two acquisitions. This registration helps experts to make an intra-patient follow-up for hepatic tumors.

Firstly, an original and efficient tree matching is applied on different segmentations of the vascular system of a single patient [1]. These vascular systems are segmented from CT-scan images acquired (every six months) during disease treatement, and then modeled as trees. Our method matches common bifurcations and vessels. Secondly, an estimation of liver deformation is computed from the results of the first step.

This approach is validated on a large synthetic database containing cases with various deformation and segmentation problems. In each case, after the registration process, the liver recovery is very accurate (around 95%) and the mean localization error for 3D landmarks in liver is small (around 4mm).

Image similarity measures for registration can be considered within the general context of joint intensity histograms, which consist of bin count parameters estimated from image intensity samples. Many approaches to estimation are ML (maximum likelihood), which tends to be unstable in the presence sparse data, resulting in registration that is driven by spurious noisy matches instead of valid intensity relationships. We propose instead a method of MAP (maximum a posteriori) estimation, which is well-defined for sparse data, or even in the absence of data. This estimator can incorporate a variety of prior assumptions, such as global histogram characteristics, or use a maximum entropy prior when no such assumptions exist. We apply our estimation method to deformable registration of MR (magnetic resonance) and US (ultrasound) images for an IGNS (image-guided guided neurosurgery) application, where our MAP estimation method results in more stable and accurate registration than a traditional ML approach.

Real-time three-dimensional ultrasound (RT3D US) is an ideal imaging modality for the diagnosis of cardiac disease. RT3D US is a flexible, inexpensive, non-invasive tool that provides important diagnostic information related to cardiac function. Unfortunately, RT3D US suffers from inherent shortcomings, such as low signal-to-noise ratio and limited field of view, producing images that are difficult to interpret. Multi-modal dynamic cardiac image registration is a well-recognized approach that compensates for these deficiencies while retaining the advantages of RT3D US imaging. The clinical application of multi-modal image registration methods is difficult, and there are a number of implementation issues to be resolved. In this work, we present a method for the rapid registration of RT3D US images of the beating heart to high-resolution magnetic resonance (MR) images. This method was validated using a volunteer image set. Validation results demonstrate that this approach can achieve rapid registration of images of the beating heart with fiducial landmark and registration errors of 1.25 ± 0.63 and 1.76 mm respectively. This technique can potentially be used to improve the diagnosis of cardiac disease by augmenting RT3D US images with high-resolution MR images and to facilitate intra-operative image fusion for minimally invasive cardio-thoracic surgical navigation.

This paper presents a learning method to select best geometric features for deformable brain registration. Best geometric features are selected for each brain location, and used to reduce the ambiguity in image matching during the deformable registration. Best geometric features are obtained by solving an energy minimization problem that requires the features of corresponding points in the training samples to be similar, and the features of a point to be different from those of nearby points. By incorporating those learned best features into the framework of HAMMER registration algorithm, we achieved about 10% improvement of accuracy in estimating the simulated deformation fields, compared to that obtained by HAMMER. Also, on real MR brain images, we found visible improvement of registration in cortical regions.

An essential goal in medical image registration is, the forward and reverse mapping matrices should be inverse to each other, i.e., inverse consistency. Conventional approaches enforce consistency in deterministic fashions, through incorporation of sub-objective cost function to impose source-destination symmetric property during the registration process. Assuming that the initial forward and reverse matching matrices have been computed and used as the inputs to our system, this paper presents a stochastic framework which yields perfect inverse consistency with the simultaneous considerations of the errors underneath the registration matrices and the imperfectness of the consistent constraint. An iterative generalized total least square (GTLS) strategy has been developed such that the inverse consistency is optimally imposed.

We propose a novel incremental surface-based registration technique that employs the Unscented Kalman Filter (UKF) to register two different data sets. The method not only reports the variance of the registration parameters but also has significantly more accurate results in comparison to the Iterative Closest Points (ICP) algorithm. Furthermore, it is shown that the proposed incremental registration algorithm is less sensitive to the initial alignment of the data sets than the ICP algorithm. We have validated the method by registering bone surfaces extracted from a set of 3D ultrasound images to the corresponding surface points gathered from the Computed Tomography (CT) data.

Dynamic contrast-enhanced 4-D MR renography has the potential for broad clinical applications, but suffers from respiratory motion that limits analysis and interpretation. Since each examination yields at least over 10-20 serial 3-D images of the abdomen, manual registration is prohibitively labor-intensive. Besides in-plane motion and translation, out-of-plane motion and rotation are observed in the image series. In this paper, a novel robust and automated technique for removing out-of-plane translation and rotation with sub-voxel accuracy in 4-D dynamic MR images is presented. The method was evaluated on simulated motion data derived directly from a clinical patient’s data. The method was also tested on 24 clinical patient kidney data sets. Registration results were compared with a mutual information method, in which differences between manually co-registered time-intensity curves and tested time-intensity curves were compared. Evaluation results showed that our method agreed well with these ground truth data.

This paper deals with the modeling of a vascular C-arm to generate 3D augmented fluoroscopic images in an interventional radiology context. A methodology based on the use of a multi-image calibration is proposed to assess the physical behavior of the C-arm. From the knowledge of the main characteristics of the C-arm, realistic models of the acquisition geometry are proposed. Their accuracy was evaluated and experiments showed that the C-arm geometry can be predicted with a mean 2D reprojection error of 0.5 mm. The interest of 3D augmented fluoroscopy is also assessed on a clinical case.

Statistical atlases built by point distribution models (PDMs) using a novel hybrid 3D shape model were used for surface reconstruction. The hybrid shape model removes the need for global scaling in aligning training examples and instance generation, thereby allowing the PDM to capture a wider range of variations. The atlases can be used to reconstruct, or deformably register, the surface model of an object from just two to four 2D x-ray projections of the object. The methods was tested using proximal and distal femurs. Results of simulated projections and fluoroscopic images of cadaver knees show that the new instances can be registered with an accuracy of about 2 mm.

In this paper we present a novel 3D/2D registration method, where first, a 3D image is reconstructed from a few 2D X-ray images and next, the preoperative 3D image is brought into the best possible spatial correspondence with the reconstructed image by optimizing a similarity measure. Because the quality of the reconstructed image is generally low, we introduce a novel asymmetric mutual information similarity measure, which is able to cope with low image quality as well as with different imaging modalities. The novel 3D/2D registration method has been evaluated using standardized evaluation methodology and publicly available 3D CT, 3DRX, and MR and 2D X-ray images of two spine phantoms [1], for which gold standard registrations were known. In terms of robustness, reliability and capture range the proposed method outperformed the gradient-based method [2] and the method based on digitally reconstructed radiographs (DRRs).

Accurate 3D/2D vessel registration is complicated by issues of image quality, occlusion, and other problems. This study performs a quantitative comparison of 3D/2D vessel registration in which vessels segmented from preoperative CT or MR are registered with biplane x-ray angiograms by either a) simultaneous two-view registration with advance calculation of the relative pose of the two views, or b) sequential registration with each view. We conclude on the basis of phantom studies that, even in the absence of image errors, simultaneous two-view registration is more accurate than sequential registration. In more complex settings, including clinical conditions, the relative accuracy of simultaneous two-view registration is even greater.

Voxel based non-rigid registration of images involves finding a similarity maximising transformation that deforms a source image to the coordinate system of a target image. In order to do this, interpolation is required to estimate the source intensity values corresponding to transformed target voxels. These interpolated source intensities are used when calculating the similarity measure being optimised. In this work, we compare the extent and nature of artefactual displacements produced by voxel based non-rigid registration techniques for different interpolators and investigate their relationship to image noise and global transformation error. A per-voxel similarity gradient is calculated and the resulting vector field is used to characterise registration artefacts for each interpolator. Finally, we show that the resulting registration artefacts can generate spurious volume changes for image pairs with no expected volume change.

The need for non-rigid multi-modal registration is becoming increasingly common for many clinical applications. To date, however, existing proposed techniques remain as largely academic research effort with very few methods being validated for clinical product use. It has been suggested by Crum et al. [1] that the context-free nature of these methods is one of the main limitations and that moving towards context-specific methods by incorporating prior knowledge of the underlying registration problem is necessary to achieve registration results that are accurate and robust enough for clinical applications. In this paper, we propose a novel non-rigid multi-modal registration method using a variational formulation that incorporates a prior learned joint intensity distribution. The registration is achieved by simultaneously minimizing the Kullback-Leibler divergence between an observed and a learned joint intensity distribution and maximizing the mutual information between reference and alignment images. We have applied our proposed method on both synthetic and real images with encouraging results.

An approach to deformable registration of three-dimensional brain tumor images to a normal brain atlas is presented. The approach involves the integration of three components: a biomechanical model of tumor mass-effect, a statistical approach to estimate the model’s parameters, and a deformable image registration method. Statistical properties of the desired deformation map are first obtained through tumor mass-effect simulations on normal brain images. This map is decomposed into the sum of two components in orthogonal subspaces, one representing inter-individual differences, and the other involving tumor-induced deformation. For a new tumor case, a partial observation of the desired deformation map is obtained via deformable image registration and is decomposed into the aforementioned spaces in order to estimate the mass-effect model parameters. Using this estimate, a simulation of tumor mass-effect is performed on the atlas to generate an image that is more similar to brain tumor image, thereby facilitating the atlas registration process. Results for a real and a simulated tumor case indicate significant reduction in the registration error due to the presented approach as compared to the direct use of deformable image registration.

Tagged Magnetic Resonance Imaging (MRI) is currently the reference MR modality for myocardial motion and strain analysis. NMI-based non rigid registration has proven to be an accurate method to retrieve cardiac deformation fields. The use of

α

MI permits higher dimensional features to be implemented in myocardial deformation estimation through image registration. This paper demonstrates that this is feasible with a set of Haar wavelet features of high dimension. While we do not demonstrate performance improvement for this set of features, there is no significant degradation as compared to implementing the registration method with the traditional NMI metric. We use Entropic Spanning Graphs (ESGs) to estimate the

α

MI of the wavelet feature vectors WFVs since this is not possible with histograms. To the best of our knowledge, this is the first time that ESGs are used for non rigid registration.

Robust 3D point registration is difficult for biomedical surfaces, especially for roundish and approximate symmetric soft tissues such as liver, stomach, etc. We present an Iterative Optimization Registration scheme (IOR) based on Hierarchical Vertex Signatures (HVS) between point-sets of medical surfaces. HVSs are distributions of concatenated neighborhood angles relative to the PCA axes of the surfaces which concisely describe global structures and local contexts around vertices in a hierarchical paradigm. The correspondences between point-sets are then established by Chi-Square test statistics. Specifically, to alleviate the sensitivity to axes directions that often affects robustness for other global axes based algorithms, IOR aligns surfaces gradually, and incrementally calibrates the directions of major axes in a multi-resolution manner. The experimental results demonstrate IOR is efficient and robust for liver registration. This method is also promising to other applications such as morphological pathological analysis, 3D model retrieval and object recognition.

Optimal port placement is a delicate issue in minimally invasive endoscopic surgery, particularly in robotically assisted surgeey. A good choice of the instruments’ and endoscope’s ports can avoid time-consuming consecutive new port placement. We present a novel method to intuitively and precisely plan the port placement. The patient is registered to its pre-operative CT by just moving the endoscope around fiducials, which are attached to the patient’s thorax and are visible in its CT. Their 3D positions are automatically reconstructed. Without prior time-consuming segmentation, the pre-operative CT volume is directly rendered with respect to the endoscope or instruments. This enables the simulation of a camera flight through the patient’s interior along the instruments’ axes to easily validate possible ports.

We present a new algorithm to register 3D pre-operative Magnetic Resonance (MR) images with intra-operative MR images of the brain. This algorithm relies on a robust estimation of the deformation from a sparse set of measured displacements. We propose a new framework to compute iteratively the displacement field starting from an approximation formulation (minimizing the sum of a regularization term and a data error term) and converging toward an interpolation formulation (least square minimization of the data error term). The robustness of the algorithm is achieved through the introduction of an outliers rejection step in this gradual registration process. We ensure the validity of the deformation by the use of a biomechanical model of the brain specific to the patient, discretized with the finite element method. The algorithm has been tested on six cases of brain tumor resection, presenting a brain shift up to 13 mm.

We present a framework for rigid registration of a set of B-mode ultrasound images to a CT scan in the context of Radiotherapy planning. Our main focus is on deriving an appropriate similarity measure based on the physical properties and artifacts of ultrasound. A combination of a weighted Mutual Information term, edge correlation, clamping to the skin surface and occlusion detection is able to assess the alignment of structures in ultrasound images and simulated slices generated from the CT data. Hence a set of ultrasound images, whose relative transformations are given by a magnetic tracking device, can be registered automatically to the CT scan. We validated our methods on neck data of patients with head and neck tumors and cervical lymph node metastases.

Four-dimensional (4D) computed tomography (CT) image acquisition is a useful technique in radiation treatment planning and interventional radiology in that it can account for respiratory motion of lungs. Current 4D lung reconstruction techniques have limitations in either spatial or temporal resolution. In addition, most of these techniques rely on auxiliary surrogates to relate the time of CT scan to the patient’s respiratory phase. In this paper, we propose a novel 4D CT lung reconstruction and deformation estimation algorithm. Our algorithm is purely image based. The algorithm can reconstruct high quality 4D images even if the original images are acquired under irregular respiratory motion. The algorithm is validated using synthetic 4D lung data. Experimental results from a swine study data are also presented.

This paper describes an automatic parameter optimization method for anisotropic diffusion filters used to de-noise 2D and 3D MR images. The filtering process is integrated into a closed-loop system where image improvement is monitored indirectly by comparing the characteristics of the suppressed noise with those of the assumed noise model at the optimal point. In order to verify the performance of this approach, experimental results obtained with this method are presented together with the results obtained by median and k-nearest neighbor filters.

We approach the problem of temporal reparameterization of dynamic sequences of lung MR images. In earlier work, we employed capacity-based reparameterization to co-register temporal sequences of 2-D coronal images of the human lungs. Here, we extend that work to the evaluation of a ventilator-acquired 3-D dataset from a normal mouse. Reparameterization according to both deformation and lung volume is evaluated. Both measures provide results that closely approximate normal physiological behavior, as judged from the original data. Our ultimate goal is to be able to characterize normal parenchymal biomechanics over a population of healthy individuals, and to use this statistical model to evaluate lung deformation under various pathological states.

Respiratory organ motion is a key problem in proton therapy and in many other treatments. This paper presents a novel retrospective gating method for 4D (dynamic 3D) MR imaging during free breathing to capture the full variability of respiratory organ deformation. In contrast to other imaging methods, a constant breathing depth or even strict periodicity are not assumed. 3D images of moving organs can be reconstructed for complete respiratory cycles by retrospective stacking of dynamic 2D images using internal image-based gating. Additional noise reduction by combining multiple images significantly increases the signal-to-noise ratio. The resulting image quality is comparable to breath-hold acquisitions. Although the method was developed for proton therapy planning, the new possibilities to study respiratory motion are valuable to improve other treatments and to assess gating techniques, which rely on stronger assumptions about the breathing pattern.

We present here a method that aims at defining a surface-based coordinate system on the cortical surface. Such a system is needed for both cortical localization and intersubject matching in the framework of neuroimaging. We propose an automatic parameterization based on the spherical topology of the grey/white matter interface of each hemisphere and on the use of naturally organized and reproducible anatomical features. From those markers used as initial constraints, the coordinate system is propagated via a PDE solved on the cortical surface.

We describe a novel multiresolution parametric framework to estimate transparent motions typically present in X-Ray exams. Assuming the presence if two transparent layers, it computes two affine velocity fields by minimizing an appropriate objective function with an incremental Gauss-Newton technique. We have designed a realistic simulation scheme of fluoroscopic image sequences to validate our method on data with ground truth and different levels of noise. An experiment on real clinical images is also reported. We then exploit this transparent-motion estimation method to denoise two layers image sequences using a motion-compensated estimation method. In accordance with theory, we show that we reach a denoising factor of 2/3 in a few iterations without bringing any local artifacts in the image sequence.

Computed tomography angiography (CTA) is an established tool for vessel imaging. Yet, high-intense structures in the contrast image can seriously hamper luminal visualisation. This can be solved by subtraction CTA, where a native image is subtracted from the contrast image. However, patient and organ motion limit the application of this technique. Within this paper, a fully automated intensity-based nonrigid 3D registration algorithm for subtraction CT angiography is presented, using a penalty term to avoid volume change during registration. Visual and automated validation on four clinical datasets clearly show that the algorithm strongly reduces motion artifacts in subtraction CTA. With our method, 39% to 99% of the artifacts disappear, also those caused by minimal displacement of stents or calcified plaques. This results in a better visualisation of the vessel lumen, also of the smaller vessels, allowing a faster and more accurate inspection of the whole vascular structure, especially in case of stenosis.

In Emission Tomography imaging, respiratory motion causes artifacts in lungs and cardiac reconstructed images, which lead to misinterpretations and imprecise diagnosis. Solutions like respiratory gating, correlated dynamic PET techniques, list-mode data based techniques and others have been tested with improvements over the spatial activity distribution in lungs lesions, but with the disadvantages of requiring additional instrumentation or discarding part of the projection data used for reconstruction. The objective of this study is to incorporate respiratory motion correction directly into the image reconstruction process, without any additional acquisition protocol consideration. To this end, we propose an extension to the Maximum Likelihood Expectation Maximization (MLEM) algorithm that includes a respiratory motion model, which takes into account the displacements and volume deformations produced by the respiratory motion during the data acquisition process. We present results from synthetic simulations incorporating real respiratory motion as well as from phantom and patient data.

Fast retrieval using organ shapes is crucial in medical image databases since shape is a clinically prominent feature. In this paper, we propose that 2-D shapes in medical image databases can be indexed by embedding them into a vector space and using efficient vector space indexing. An optimal shape space embedding is proposed for this purpose. Experimental results of indexing vertebral shapes in the NHANES II database are presented. The results show that vector space indexing following embedding gives superior performance than metric indexing.

In functional neurosurgery, there is a growing need for accurate localization of the functional targets. Since deep brain stimulation (DBS) of the Vim thalamic nucleus has been proposed for the treatment of Parkinson’s disease, the target has evolved toward the globus pallidus and subthalamic nucleus (STN) and the therapeutic indications have enlarged to include psychiatric disorders such as Tourette syndrome or obsessive compulsive disorders. In these pathologies, the target has been restrained to smaller functional subterritories of the basal ganglia, requiring more refined techniques to localize smaller and smaller brain regions, often invisible in routine clinical MRI. Different strategies have been developed to identify such deep brain targets. Direct methods can identify structures in the MRI itself, but only the larger ones. Indirect methods are based on the use of anatomical atlases. The present strategy comprised a 3D histological atlas and the MRI of the same brain specimen, and deformation methodology developped to fit the atlas toward the brain of any given patient. In this paper, this method is evaluated in the aim of being applied to further studies of anatomo-clinical correlation. The accuracy of the method is first discussed, followed by the study of short series of Parkinsonian patients treated by DBS, allowing to compare the deformed atlas with various per- and post-operative data.

Digital brain atlases can be used in conjunction with magnetic resonance imaging (MRI) and computed tomography (CT) for planning and guidance during neurosurgery. Digital atlases are advantageous, since they can be warped nonlinearly to fit each patient’s unique anatomy.

Two atlas-to-patient warping techniques are compared in this paper. The first technique uses an MRI template as an intermediary to estimate a nonlinear atlas-to-patient transformation. The second, is novel, and uses a pseudo-MRI volume, derived from the voxel-label-atlas, to estimate the atlas-to-patient transformation directly. Manual segmentations and functional data are used to validate the two methods.

In this paper we present a novel method for building a 4D statistical atlas describing the cardiac anatomy and how the cardiac anatomy changes during the cardiac cycle. The method divides the distribution space of cardiac shapes into two subspaces. One distribution subspace accounts for changes in cardiac shape caused by inter-subject variability. The second distribution subspace accounts for changes in cardiac shape caused by deformation during the cardiac cycle (i.e. intra-subject variability). Principal component analysis (PCA) have been performed in order to calculate the most significant modes of variation of each distribution subspace. During the construction of the statistical atlas we eliminate the need for manual landmarking of the cardiac images by using a non-rigid surface registration algorithm to propagate a set of pseudo-landmarks from an automatically landmarked atlas to each frame of all the image sequences. In order to build the atlas we have used 26 cardiac image sequences from healthy volunteers. We show how the resulting statistical atlas can be used to differentiate between cardiac image sequences from patients with hypertrophic cardiomyopathy and normal subjects.

The construction of population atlases is a key issue in medical image analysis, and particularly in brain mapping. Large sets of images are mapped into a common coordinate system to study intra-population variability and inter-population differences, to provide voxel-wise mapping of functional sites, and to facilitate tissue and object segmentation via registration of anatomical labels. We formulate the unbiased atlas construction problem as a Fréchet mean estimation in the space of diffeomorphisms via large deformations metric mapping. A novel method for computing constant speed velocity fields and an analysis of atlas stability and robustness using entropy are presented. We address the question: how many images are required to build a stable brain atlas?

Probabilistic atlas has broad applications in medical image segmentation and registration. The most common problem building a probabilistic atlas is picking a target image upon which to map the rest of the training images. Here we present a method to choose a target image that is the closest to the mean geometry of the population under consideration as determined by bending energy. Our approach is based on forming a distance matrix based on bending energies of all pair-wise registrations and performing multidimensional scaling (MDS) on the distance matrix.

In this paper we study and evaluate the influence of the choice of a particular reference volume as the electrophysiological atlas on the accuracy of the automatic predictions of optimal points for deep brain stimulator (DBS) implants. We refer to an electrophysiological atlas as a spatial map of electrophysiological information such as micro electrode recordings (MER), stimulation parameters, final implants positions, etc., which are acquired for each patient and then mapped onto a single reference volume using registration algorithms. An atlas-based prediction of the optimal point for a DBS surgery is made by registering a patient’s image volume to that reference volume, that is, by computing a correct coordinate mapping between the two; and then by projecting the optimal point from the atlas to the patient using the transformation from the registration algorithm. Different atlases, as well as different parameterizations of the registration algorithm, lead to different and somewhat independent atlas-based predictions. We show how the use of multiple reference volumes can improve the accuracy of prediction by combining the predictions from the multiple reference volumes weighted by the accuracy of the non-rigid registration between each of the corresponding atlases and the patient volume.

The growing usage of statistical shape analysis in medical imaging calls for effective methods for highly accurate shape correspondence. This paper presents a novel landmark-based method to correspond a set of 2D shape instances in a nonrigid fashion. Different from prior methods, the proposed method combines three important factors in measuring the shape-correspondence error: landmark-correspondence error, shape-representation error, and shape-representation compactness. In this method, these three important factors are explicitly handled by the landmark sliding, insertion, and deletion operations, respectively. The proposed method is tested on several sets of structural shape instances extracted from medical images. We also conduct an empirical study to compare the developed method to the popular Minimum Description Length method.

This paper describes a technique to approximately predict the facial morphology after standardized orthognathic ostoetomies. The technique only relies on the outer facial morphology represented as a set of surface points and does not require computed tomography (CT) images as input. Surface points may either be taken from 3D surface scans or from 3D positions palpated on the face using a tracking system. The method is based on a statistical model generated from a set of pre- and postoperative 3D surface scans of patients that underwent the same standardized surgery. The model contains both the variability of preoperative facial morphologies and the corresponding postoperative deformations. After fitting the preoperative part to 3D data from a new patient the preoperative face is approximated by the model and the prediction of the postoperative morphology can be extracted at the same time. We built a model based on a set of 15 patient data sets and tested the predictive power in leave-one-out tests for a set of relevant cephalometric landmarks. The average prediction error was found to be between 0.3 and 1.2 mm at all important facial landmarks in the relevant areas of upper and lower jaw. Thus the technique provides an easy and powerful way of prediction which avoids time, cost and radiation required by other prediction techniques such as those based on CT scans.

In this paper, we present a new framework for shape modelling and analysis: we suggest to look at the problem from a pattern recognition point of view, and claim that under this prospective several advantages are achieved. The modelling of a surface with a point distribution model is seen as an unsupervised clustering problem, and tackled by using growing cell structures. The adaptation of a model to new shapes is studied as a classification task, and provides a straightforward solution to the point correspondence problem in active shape modelling. The method is illustrated and tested in 3D synthetic datasets and applied to the modelling of brain ventricles in an elderly population.

Shape priors attempt to represent biological variations within a population. When variations are global, Principal Component Analysis (PCA) can be used to learn major modes of variation, even from a limited training set. However, when significant local variations exist, PCA typically cannot represent such variations from a small training set. To address this issue, we present a novel algorithm that learns shape variations from data at

multiple scales and locations

using spherical wavelets and spectral graph partitioning. Our results show that when the training set is small, our algorithm significantly improves the approximation of shapes in a testing set over PCA, which tends to oversmooth data.

In this work, a discriminative model of attention deficit hyperactivity disorder (ADHD) is presented on the basis of multivariate pattern classification and functional magnetic resonance imaging (fMRI). This model consists of two parts, a classifier and an intuitive representation of discriminative pattern of brain function between patients and normal controls. Regional homogeneity (ReHo), a measure of brain function at resting-state, is used here as a feature of classification. Fisher discriminative analysis (FDA) is performed on the features of training samples and a linear classifier is generated. Our initial experimental results show a successful classification rate of 85%, using leave-one-out cross validation. The classifier is also compared with linear support vector machine (SVM) and Batch Perceptron. Our classifier outperforms the alternatives significantly. Fisher brain, the optimal projective-direction vector in FDA, is used to represent the discriminative pattern. Some abnormal brain regions identified by Fisher brain, like prefrontal cortex and anterior cingulate cortex, are well consistent with that reported in neuroimaging studies on ADHD. Moreover, some less reported but highly discriminative regions are also identified. We conclude that the discriminative model has potential ability to improve current diagnosis and treatment evaluation of ADHD.

FMRI group studies are usually based on stereotactic spatial normalization and present voxel by voxel average activity across subjects. This technique does not in general adequately model the inter subject spatial variability. In this work, we propose to identify functional landmarks that are reliable across subjects with subject specific Talairach coordinates that are similar -but not exactly identical- between subjects. We call these Brain Functional Landmarks (BFLs), and define them based on cross-validation techniques using 38 subjects. We explore a dataset acquired while subjects were involved in several cognitive and sensori-motor processes, and show that this representation allows to classify subjects into sub-groups on the basis of their BFL activity.

We present here a new method for correcting the topology of objects segmented from medical images. Whereas previous techniques alter a surface obtained from the hard segmentation of the object, our technique works directly in the image domain, propagating the topology for all isosurfaces of the object. From an analysis of topological changes and critical points in implicit surfaces, we introduce a topology progagation algorithm that enforces any desired topology using a fast marching technique. Compared to previous topology correction techniques, the method successfully corrects topology while effecting fewer changes to the original volume.

We present a method for the analysis of deep grey brain nuclei for accurate detection of human spongiform encephalopathy in multisequence MRI of the brain. We employ T1, T2 and FLAIR-T2 MR sequences for the detection of intensity deviations in the internal nuclei. The MR data are registered to a probabilistic atlas and normalised in intensity prior to the segmentation of hyperintensities using a foveal model. Anatomical data from a segmented atlas are employed to refine the registration and remove false positives. The results are robust over the patient data and in accordance to the clinical ground truth. Our method further allows the quantification of intensity distributions in basal ganglia. sCJD patient FLAIR images are classified with a more significant hypersignal in caudate nuclei (10/10) and putamen (6/10) than in thalami. Defining normalised MRI measures of the intensity relations between the internal grey nuclei of patients, we robustly differentiate sCJD and variant CJD (vCJD) patients, as an attempt towards the automatic detection and classification of human spongiform encephalopathies.

This paper proposes an approach to effectively representing the statistics of high-dimensional deformations, when relatively few training samples are available, and conventional methods, like PCA, fail due to insufficient training. Based on previous work on scale-space decomposition of deformation fields, herein we represent the space of “valid deformations” as the intersection of three subspaces: one that satisfies constraints on deformations themselves, one that satisfies constraints on Jacobian determinants of deformations, and one that represents smooth deformations via a Markov Random Field (MRF). The first two are extensions of PCA-based statistical shape models. They are based on a wavelet packet basis decomposition that allows for more accurate estimation of the covariance structure of deformation or Jacobian fields, and they are used jointly due to their complementary strengths and limitations. The third is a nested MRF regularization aiming at eliminating potential discontinuities introduced by assumptions in the statistical models. A randomly sampled deformation field is projected onto the space of valid deformations via iterative projections on each of these subspaces until convergence,

i.e.

all three constraints are met. A deformation field simulator uses this process to generate random samples of deformation fields that are not only realistic but also representative of the full range of anatomical variability. These simulated deformations can be used for validation of deformable registration methods. Other potential uses of this approach include representation of shape priors in statistical shape models as well as various estimation and hypothesis testing paradigms in the general fields of computational anatomy and pattern recognition.

We propose a novel method for quantitative interpretation of uncalibrated optical images which is derived explicitly from an analysis of the image formation model. Parameters characterising the tissue are recovered from images acquired using filters optimised to minimise the error. Preliminary results are shown for the skin, where the technique was successfully applied to aid the diagnosis and interpretation of non-melanocytic skin cancers and acne; and for the more challenging ocular fundus, for mapping of the pigment xanthophyll.

We introduce a novel approach for magnetic resonance image (MRI) brain tissue classification by learning image neighborhood statistics from noisy input data using nonparametric density estimation. The method models images as random fields and relies on minimizing an entropy-based metric defined on high dimensional probability density functions. Combined with an atlas-based initialization, it is completely automatic. Experiments on real and simulated data demonstrate the advantages of the method in comparison to other approaches.

The minimally invasive treatment of liver tumors represents an alternative to the open surgery approach. Radio-frequency ablation destroys a tumor by delivering radio-frequency energy through a needle probe. Traditionally, the probe is placed manually using imaging feedback. New approaches use robotic devices to accurately place the instrument at the target. The authors developed an image-guided robotic system for percutaneous interventions using computed tomography. The paper presents a randomized patient study comparing the manual versus robotic needle placement for radio-frequency ablation procedures of liver tumors. The results of this study show that in our case robotic interventions were a very viable solution. Several treatment parameters such as radiation exposures and procedure-times were found to be significantly improved in the robotic case.

For over 20 years, interventional methods have improved the outcomes of patients with cardiovascular disease. However, these procedures require an intricate combination of visual and tactile feedback and extensive training periods. In this paper, we describe a series of novel approaches that have lead to the development of a high-fidelity simulation system for interventional neuroradiology. In particular we focus on a new approach for real-time deformation of devices such as catheters and guidewires during navigation inside complex vascular networks. This approach combines a real-time incremental Finite Element Model, an optimization strategy based on substructure decomposition, and a new method for handling collision response in situations where the number of contacts points is very large. We also briefly describe other aspects of the simulation system, from patient-specific segmentation to the simulation of contrast agent propagation and fast volume rendering techniques for generating synthetic X-ray images in real-time.

In this paper, we propose a hybrid method for tracking a bronchoscope that uses a combination of magnetic sensor tracking and image registration. The position of a magnetic sensor placed in the working channel of the bronchoscope is provided by a magnetic tracking system. Because of respiratory motion, the magnetic sensor provides only the approximate position and orientation of the bronchoscope in the coordinate system of a CT image acquired before the examination. The sensor position and orientation is used as the starting point for an intensity-based registration between real bronchoscopic video images and virtual bronchoscopic images generated from the CT image. The output transformation of the image registration process is the position and orientation of the bronchoscope in the CT image. We tested the proposed method using a bronchial phantom model. Virtual breathing motion was generated to simulate respiratory motion. The proposed hybrid method successfully tracked the bronchoscope at a rate of approximately 1 Hz.

Active robotic filtering is a promising solution for beating heart Totally Endoscopic Coronary Artery Bypass Grafting (TECABG). In this work, we assess the heart motion dynamics using simultaneously high speed imaging of optical markers attached to the heart, ECG signals and ventilator airflow acquisitions. Our goal is to make an assessment of the heart motion (shape, velocity, acceleration) in order to be able to make more accurate specifications for a dedicated robot that could follow this motion in real-time. Furthermore, using the 2 additional inputs (ECG, airflow), we propose a prediction algorithm of the motion that could be used with a predictive control algorithm to improve the tracking accuracy.

In this study we have designed and created a data-fusion display that has enabled volumetric MIP image navigation using intraoperative C-arm CT data in the operating room. The 3D volumetric data reflecting a patient’s inner structure is directly displayed on the monitor through video images of the surgical field using a 3D optical tracking system, a ceiling-mounted articulating monitor, and a small size video camera mounted at the back of the monitor. The system performance was validated in an experiment carried out in the operating room.

An integral element of every surgical simulator is the ability to interactively cut tissue. A number of approaches have been suggested in the past, the most important being mesh subdivision by introducing new elements and mesh adaptation by adjusting existing topology. In this paper we combine these two methods and optimize them for our training system of hysteroscopic interventions. The basic methodology is introduced in 2D, a first extension to 3D is presented and finally the integration into the simulator described.

During hysteroscopy a hydrometra is maintained, i.e. the uterus is distended with liquid media to access and visualize the uterine cavity. The pressure and flow induced by the liquid are crucial tools for the gynecologists during surgery to obtain a clear view of the operation site. This paper presents two different aspects of hydrometra simulation, namely the distension of the uterine muscle and the liquid flow simulation in the cavity. The deformation of the organ’s shape is computed offline based on finite element calculations whereas the flow is approximated on the fly by solving the simplified Navier-Stokes equations. The real-time capabilities of the presented algorithms as well as the level of fidelity achieved by the proposed methods are discussed.

In the present study, fully nonlinear (i.e. accounting for both geometric and material nonlinearities) patient specific finite element brain model was applied to predict deformation field within the brain during the craniotomy-induced brain shift. Deformation of brain surface was used as displacement boundary conditions. Application of the computed deformation field to align (i.e. register) the preoperative images with the intraoperative ones indicated that the model very accurately predicts the displacements of gravity centers of the lateral ventricles and tumor even for very limited information about the brain surface deformation. These results are sufficient to suggest that nonlinear biomechanical models can be regarded as one possible way of complementing medical image processing techniques when conducting nonrigid registration. Important advantage of such models over the linear ones is that they do not require unrealistic assumptions that brain deformations are infinitesimally small and brain tissue stress–strain relationship is linear.

Cornea surgeons have observed that changes in cornea curvature can follow cataract surgery and cause astigmatism. The placement of surgical incisions has been shown to influence these curvature changes. Though empirical data has been collected about this phenomenon, a biomechanical model has not been employed in predicting post-surgical outcomes. This work implemented an incised finite element model of the eye to investigate factors influencing corneal shape after surgery. In particular, the effects of eye muscle forces and intra-ocular pressure were simulated. Cornea shape change was computed via finite element analysis, and the resulting change in cornea curvature was measured by fitting quadratic curves to the horizontal and vertical meridians of the cornea. Results suggest that these two sources of deforming force counteract each other and contribute to astigmatism in perpendicular directions.

Soft tissue characterization and modeling based on living tissues has been investigated in order to provide a more realistic behavior in a virtual reality based surgical simulation. In this paper, we characterize the nonlinear viscoelastic properties of intra-abdominal organs using the data from

in vivo

animal experiments and inverse FE parameter estimation algorithm. In the assumptions of quasi-linear-viscoelastic theory, we estimated the viscoelastic and hyperelastic material parameters to provide a physically based simulation of tissue deformations. To calibrate the parameters to the experimental results, we developed a three dimensional FE model to simulate the forces at the indenter and an optimization program that updates new parameters and runs the simulation iteratively. We can successfully reduce the time and computation resources by decoupling the viscoelastic part and nonlinear elastic part in a tissue model. The comparison between simulation and experimental behavior of pig intra abdominal soft tissue are presented to provide a validness of the tissue model using our approach.

Cardiac ablation procedures are becoming more routine to treat arrhythmias. The development of electrophysiological models will allow investigation of treatment strategies. However, current models are computationally expensive and often too complex to be adjusted with current clinical data. In this paper, we have proposed a fast algorithm to solve Eikonal-based models on triangular meshes. These models can be used to extract hidden parameters of the cardiac function from clinical data in a very short time, thus could be used during interventions. We propose a first approach to estimate these parameters, and have tested it on synthetic and real data derived using XMR imaging. We demonstrated a qualitative matching between the estimated parameter and XMR data. This novel approach opens up possibilities to directly integrate modelling in the interventional room.

We present a new technique for determining structure-by-structure volume changes, using an inverse problem approach. Given a pre-labelled brain and a series of images at different time-points, we generate finite element meshes from the image data, with volume change modelled by means of an unknown coefficient of expansion on a per-structure basis. We can then determine the volume change in each structure of interest using inverse problem optimization techniques. The proposed method has been tested with simulated and clinical data. Results suggest that the presented technique can be seen as an alternative for volume change estimation.

Models that predict the soft tissue deformation caused by needle insertion could improve the accuracy of procedures such as brachytherapy and needle biopsy. Prior work on needle insertion modeling has focused on static deformation; the experiments presented here show that dynamic effects such as relaxation are important. An experimental setup is described for recording and measuring the deformation that occurs with needle insertion into a soft tissue phantom. Analysis of the collected data demonstrates the time- and velocity-dependent nature of the deformation. Deformation during insertion is shown to be well represented using a velocity-dependent force function with a linear elastic finite element model. The model’s accuracy is limited to the period during needle motion, indicating that a viscoelastic tissue model may be required to capture tissue relaxation after the needle stops.

In order to calculate deformation of soft tissue under arbitrary loading conditions, we have to take both non-linear material characteristics and subcutaneous structures into considerations. The estimation method of material properties presented in this paper accounts for these issues. It employs a compression test inside MRI in order to visualize deformation of hypodermic layered structure of living tissue, and an FE model of the compressed tissue in which non-linear material model is assigned. The FE analysis is iterated with updated material constant until the difference between the displacement field observed from MR images and calculated by FEM is minimized. The presented method has been applied to a 3-layered silicon rubber phantom. The results show the excellent performance of our method. The accuracy of the estimation is better than 15 %, and the reproducibility of the deformation is better than 0.4 mm even for an FE analysis with different boundary condition.

Better physiological understanding of principles regulating vascular formation and growth is mandatory to their efficient modeling for the purpose of physiologically oriented medical applications like training simulation or pre-operative planning. We have already reported on the implementation of a visually oriented modeling framework allowing to study various physiological aspects of the vascular systems on a macroscopic scale. In this work we describe our progress in this field including (i) extension of the presented model to three dimensions, (ii) addition of established mathematical approaches to modeling angiogenesis and (iii) embedding the structures in arbitrary anatomical elements represented by finite element meshes.

Diffuse optical tomography (DOT) is a noninvasive imaging technology that is sensitive to local concentration changes in oxy- and deoxyhemoglobin. When applied to functional neuroimaging, DOT measures hemodynamics in the scalp and brain that reflect competing metabolic demands and cardiovascular dynamics. Separating the effects of systemic cardiovascular regulation from the local dynamics is vitally important in DOT analysis. In this paper, we use auxiliary physiological measurements such as blood pressure and heart rate within a Kalman filter framework to model physiological components in DOT. We validate the method on data from a human subject with simulated local hemodynamic responses added to the baseline physiology. The proposed method significantly improved estimates of the local hemodynamics in this test case. Cardiovascular dynamics also affect the blood oxygen dependent (BOLD) signal in functional magnetic resonance imaging (fMRI). This Kalman filter framework for DOT may be adapted for BOLD fMRI analysis and multimodal studies.

We develop a general approach that uses holomorphic 1-forms to parameterize anatomical surfaces with complex (possibly branching) topology. Rather than evolve the surface geometry to a plane or sphere, we instead use the fact that all orientable surfaces are Riemann surfaces and admit conformal structures, which induce special curvilinear coordinate systems on the surfaces. Based on Riemann surface structure, we can then canonically partition the surface into patches. Each of these patches can be conformally mapped to a parallelogram. The resulting surface subdivision and the parameterizations of the components are intrinsic and stable. To illustrate the technique, we computed conformal structures for several types of anatomical surfaces in MRI scans of the brain, including the cortex, hippocampus, and lateral ventricles. We found that the resulting parameterizations were consistent across subjects, even for branching structures such as the ventricles, which are otherwise difficult to parameterize. Compared with other variational approaches based on surface inflation, our technique works on surfaces with arbitrary complexity while guaranteeing minimal distortion in the parameterization. It also offers a way to explicitly match landmark curves in anatomical surfaces such as the cortex, providing a surface-based framework to compare anatomy statistically and to generate grids on surfaces for PDE-based signal processing.

Many medical imaging applications require the computation of dense correspondence vector fields that match one surface with another. To avoid the need for a large set of manually-defined landmarks to constrain these surface correspondences, we developed an algorithm to automate the matching of surface features. It extends the mutual information method to automatically match general 3D surfaces (including surfaces with a branching topology). First, we use holomorphic 1-forms to induce consistent conformal grids on both surfaces. High genus surfaces are mapped to a set of rectangles in the Euclidean plane, and closed genus-zero surfaces are mapped to the sphere. Mutual information is used as a cost functional to drive a fluid flow in the parameter domain that optimally aligns stable geometric features (mean curvature and the conformal factor) in the 2D parameter domains. A diffeomorphic surface-to-surface mapping is then recovered that matches anatomy in 3D. We also present a spectral method that ensures that the grids induced on the target surface remain conformal when pulled through the correspondence field. Using the chain rule, we express the gradient of the mutual information between surfaces in the conformal basis of the source surface. This finite-dimensional linear space generates all conformal reparameterizations of the surface. We apply the method to hippocampal surface registration, a key step in subcortical shape analysis in Alzheimer’s disease and schizophrenia.

To compare and integrate brain data, data from multiple subjects are typically mapped into a canonical space. One method to do this is to conformally map cortical surfaces to the sphere. It is well known that any genus zero Riemann surface can be conformally mapped to a sphere. Therefore, conformal mapping offers a convenient method to parameterize cortical surfaces without angular distortion, generating an orthogonal grid on the cortex that locally preserves the metric. To compare cortical surfaces more effectively, it is advantageous to adjust the conformal parameterizations to match consistent anatomical features across subjects. This matching of cortical patterns improves the alignment of data across subjects, although it is more challenging to create a consistent conformal (orthogonal) parameterization of anatomy across subjects when landmarks are constrained to lie at specific locations in the spherical parameter space. Here we propose a new method, based on a new energy functional, to optimize the conformal parameterization of cortical surfaces by using landmarks. Experimental results on a dataset of 40 brain hemispheres showed that the landmark mismatch energy can be greatly reduced while effectively preserving conformality. The key advantage of this conformal parameterization approach is that any local adjustments of the mapping to match landmarks do not affect the conformality of the mapping significantly. We also examined how the parameterization changes with different weighting factors. As expected, the landmark matching error can be reduced if it is more heavily penalized, but conformality is progressively reduced.

We developed a new method for absolute quantification of targeted radiotracers uptake in the myocardium using hybrid SPECT/CT and an external reference point source. A segmentation algorithm based on the level set was developed to determine the endocardial edges from CT, which were subsequently applied to the physically co-registered SPECT. A 3-D Gaussian fitting method was applied for quantification of the external point source. The total targeted radiotracer activity in the myocardium was normalized to that in the point source to calculate the absolute uptake of targeted radiotracer in the myocardium. Preliminary validation was performed in rats with ischemia-induced angiogenesis. The quantified in vivo radiotracer uptake was compared to the postmortem tissue radioactive well-counting of the myocardium. Our methods worked well for identification of the endocardial edges. Quantification of the focal uptake was consistent with the well-counting data. Our methods may have the potential of providing precise absolute quantification of targeted radiotracer uptake in the myocardium.

Cine-DEMR is a new cardiac imaging technique which combines aspects of Cine and Delayed Enhancement MR. Like Cine, it displays the heart beating over time allowing for the detection of motion abnormalities. Like DEMR, non-viable (dead) tissues appear with increased signal intensity (it has been shown that the extent of non-viable tissue in the left ventricle (LV) of the heart is a direct indicator of patient survival rate). We present a technique for tracking the myocardial borders in this modality and classifying myocardial pixels as viable or non-viable. Tracking is performed using an affine deformed template of borders manually drawn on the first phase of the series and refined using an ASM-like approach. Classification employs a Support Vector Machine trained on DEMR data. We applied our technique on 75 images culled from 5 patient data sets.

X-ray microCT (computed tomography) has become a valuable tool in the analysis of vascular architecture in small animals. Because of its high resolution, a detailed assessment of blood vessel physiology and pathology is possible. Vascular measurement from noninvasive imaging is important for the study and quantification of vessel disease and can aid in diagnosis, as well as measure disease progression and response to therapy. The analysis of tracked vessel trajectories enables the derivation of vessel connectivity information, lengths between vessel junctions as well as level of ramification, contributing to a quantitative analysis of vessel architecture. In this paper, we introduce a new vessel tracking methodology based on wave propagation in oriented domains. Vessel orientation and vessel likelihood are estimated based on an eigenanalysis of gray-level Hessian matrices computed at multiple scales. An anisotropic wavefront then propagates through this vector field with a speed modulated by the maximum vesselness response at each location. Putative vessel trajectories can be found by tracing the characteristics of the propagation solution between different points. We present preliminary results from both synthetic and mouse microCT image data.

This paper describes a novel approach to automatically recover accurate correspondence over various shapes. In order to detect the features points with the capability in capturing the characteristics of an individual shape, we propose to calculate the skeletal representation for the shape curve through the medial axis transform. Employing this shape descriptor, mathematical landmarks are automatically identified based on the local feature size function, which embodies the geometric and topological information of the boundary. Before matching the resulting landmarks, shape correspondence is first approached by matching the major components of the shape curves using skeleton features. This helps in keeping the consecutive order and reducing the search space during the matching process. Point matching is then performed within each pair of corresponding components by solving a consecutive assignment problem. The effectiveness of this approach is demonstrated through experimental results on several different training sets of biomedical object shapes.

Automatic image segmentation methods often involve errors, requiring the assistance of the user to correct them. In this paper, a computer-aided design system is introduced for correcting such errors. The proposed system approximates each 3-D region by a parametric surface. Region voxels are first parametrized spherically using a coarse-to-fine subdivision method. By using the voxel positions and their parameter coordinates, control points of a rational Gaussian surface are determined through a least-squares method to approximate the region. Finally, this surface is overlaid with the volumetric image and by locally pulling or pushing it with the mouse while viewing image information, the surface is revised as needed. Typically, a few minutes are sufficient to correct errors in a region.

We describe a novel approach to combining shape and appearance features in the statistical analysis of structures in medical images. The continuous medial representation is used to relate these two types of features meaningfully. The representation imposes a shape-based coordinate system on structure interiors, in a way that uses the boundary normal as one of the coordinate axes, while providing an onto and nearly one-to-one parametrization. This coordinate system is used to sample image intensities in the context of shape. The approach is illustrated by the principal components analysis of the shape and appearance of the hippocampus in T1-weighted MRI from a schizophrenia study.

The shape and appearance of vertebrae on lateral dual x-ray absorptiometry (DXA) scans were statistically modelled. The spine was modelled by a sequence of overlapping triplets of vertebrae, using Active Appearance Models (AAMs). To automate vertebral morphometry, the sequence of trained models was matched to previously unseen scans. The dataset includes a significant number of pathologies. A new dynamic ordering algorithm was assessed for the model fitting sequence, using the best quality of fit achieved by multiple sub-model candidates. The accuracy of the search was improved by dynamically imposing the best quality candidate first. The results confirm the feasibility of substantially automating vertebral morphometry measurements even with fractures or noisy images.

While geometric deformable models have brought tremendous impacts on shape representation and analysis in medical image analysis, some of the remaining problems include the handling of boundary leakage and the lack of global understanding of boundaries. We present a modification to the geodesic active contour framework such that in.uence from local neighbors of a front point is explicitly incorporated, and it is thus capable of robustly dealing with the boundary leakage problem. The fundamental power of this strategy rests with the local integration of evolution forces for each front point within its local in.uence domain, which is adaptively determined by the local level set geometry and image/ prior information. Due to the combined e.ects of internal and external constraints on a point and the interactions with those of its neighbors, our method allows stable boundary detection when the edge information is noisy and possibly discontinuous (e.g. gaps in the boundaries) while maintaining the abilities to handle topological changes, thanks to the level set implementation. The algorithm has been implemented using the meshfree particle domain representation, and experimental results on synthetic and real images demonstrate its superior performance.

Because of the complex shape of human cortical gyri and great variation between individuals, development of effective representation schemes which allow establishment of correspondence between individuals, extraction of average structure of a population, and coregistration has proved very difficult. We introduce an approach which extracts line representations of gyri at different depths from high resolution MRI, labels main gyri semi-automatically, and extracts a template from a population using non-linear principal component analysis. The method has been tested on data from 96 healthy human volunteers. The model captures the most salient shape features of all major cortical gyri, and can be used for inter-subject registration, for investigating regionalized inter-subject variability, and for inter-hemispheric comparisons.

We propose a nonlinear statistical shape model for level set segmentation which can be efficiently implemented. Given a set of training shapes, we perform a kernel density estimation in the low dimensional subspace spanned by the training shapes. In this way, we are able to combine an accurate model of the statistical shape distribution with efficient optimization in a finite-dimensional subspace. In a Bayesian inference framework, we integrate the nonlinear shape model with a nonparametric intensity model and a set of pose parameters which are estimated in a more direct data-driven manner than in previously proposed level set methods. Quantitative results show superior performance (regarding runtime and segmentation accuracy) of the proposed nonparametric shape prior over existing approaches.

Statistical shape analysis has become of increasing interest to the neuroimaging community due to its potential to locate morphological changes. In this paper, we present the a novel combination of shape analysis and Diffusion Tensor Image (DTI) Tractography to the computation of a probabilistic, model based corpus callosum (CC) subdivision. The probabilistic subdivision is based on the distances of arc-length parameterized corpus callosum contour points to trans-callosal DTI fibers associated with an automatic lobe subdivision. Our proposed subdivision method is automatic and reproducible. Its results are more stable than the Witelson subdivision scheme or other commonly applied schemes based on the CC bounding box. We present the application of our subdivision method to a small scale study of regional CC area growth in healthy subjects from age 2 to 4 years.

A new approach to interactive segmentation based on random walks was recently introduced that shows promise for allowing physicians more flexibility to segment arbitrary objects in an image. This report has two goals: To introduce a novel computational method for applying the random walker algorithm in 2D/3D using the Graphics Processing Unit (GPU) and to provide quantitative validation studies of this algorithm relative to different targets, imaging modalities and interaction strategies.

Two novel methods are proposed for robust segmentation of pulmonary nodules in CT images. The proposed solutions locate and segment a nodule in a semi-automatic fashion with a marker indicating the target. The solutions are motivated for handling the difficulty to segment juxtapleural, or wall-attached, nodules by using only local information without a global lung segmentation. They are realized as extensions of the recently proposed robust Gaussian fitting approach. Algorithms based on i) 3D morphological opening with anisotropic structuring element and ii) extended mean shift with a Gaussian repelling prior are presented. They are empirically compared against the robust Gaussian fitting solution by using a large clinical high-resolution CT dataset. The results show 8% increase, resulting in 95% correct segmentation rate for the dataset.

We present an automated algorithm for tissue segmentation of noisy, low contrast magnetic resonance (MR) images of the brain. We use a mixture model composed of a large number of Gaussians, with each brain tissue represented by a large number of the Gaussian components in order to capture the complex tissue spatial layout. The intensity of a tissue is considered a global feature and is incorporated into the model through parameter tying of all the related Gaussians. The EM algorithm is utilized to learn the parameter-tied Gaussian mixture model. A new initialization method is applied to guarantee the convergence of the EM algorithm to the global maximum likelihood. Segmentation of the brain image is achieved by the affiliation of each voxel to a selected tissue class. The presented algorithm is used to segment 3D, T1–weighted, simulated and real MR images of the brain into three different tissues, under varying noise conditions. Quantitative results are presented and compared with state–of–the–art results reported in the literature.

This paper presents a method to extract heart structures from CTA and MRA data sets, in particular the left atrium. First, the segmented blood pool is subdivided at narrowings in small components. Second, these basic components are merged automatically so that they represent the different heart structures. The resulting cutting surfaces have a relatively small diameter compared to the diameter of the neighboring heart chambers. Both steps are controlled by only one fixed parameter. The method is fast and allows interactive post-processing by the user. Experiments on various data sets show the accuracy, robustness and repeatability of this approach.

We present a novel technique for the automatic formation of vascular trees from segmented tubular structures. Our method combines a minimum spanning tree algorithm with a minimization criterion of the Mahalanobis distance. First, a multivariate class of connected junctions is defined using a set of trained vascular trees and their corresponding image volumes. Second, a minimum spanning tree algorithm forms the tree using the Mahalanobis distance of each connection from the “connected” class as a cost function. Our technique allows for the best combination of the discrimination criteria between connected and non-connected junctions and is also modality, organ and segmentation specific.

This paper considers the problem of tissue classification in 3D MRI. More specifically, a new set of texture features, based on phase information, is used to perform the segmentation of the bones of the knee. The phase information provides a very good discrimination between the bone and the surrounding tissues, but is usually not used due to phase unwrapping problems. We present a method to extract textural information from the phase that does not require phase unwrapping. The textural information extracted from the magnitude and the phase can be combined to perform tissue classification, and used to initialise an active shape model, leading to a more precise segmentation.

A novel shape based segmentation approach is proposed by modifying the external energy component of a deformable model. The proposed external energy component depends not only on the gray level of the images but also on the shape information which is obtained from the signed distance maps of objects in a given data set. The gray level distribution and the signed distance map of the points inside and outside the object of interest are accurately estimated by modelling the empirical density function with a linear combination of discrete Gaussians (LCDG) with positive and negative components. Experimental results on the segmentation of the kidneys from low-contrast DCE-MRI and on the segmentation of the ventricles from brain MRI’s show how the approach is accurate in segmenting 2-D and 3-D data sets. The 2D results for the kidney segmentation have been validated by a radiologist and the 3D results of the ventricle segmentation have been validated with a geometrical phantom.

Preventing complications during hepatic surgery in living-donor transplantation or in oncologic resections requires a careful preoperative analysis of the hepatic venous anatomy. Such an analysis relies on CT hepatic venography data, which enhances the vascular structure due to contrast medium injection. However, a 3D investigation of the enhanced vascular anatomy based on typical computer vision tools is ineffective because of the large amount of occlusive opacities to be removed. This paper proposes an automated 3D approach for the segmentation of the vascular structure in CT hepatic venography, providing the appropriate tools for such an investigation. The developed methodology relies on advanced topological and morphological operators applied in mono- and multiresolution filtering schemes. It allows to discriminate the opacified vessels from the bone structures and liver parenchyma regardless of noise presence or inter-patient variability in contrast medium dispersion. The proposed approach was demonstrated at different phases of hepatic perfusion and is currently under extensive validation in clinical routine.

Combination of multiple segmentations has recently been introduced as an effective method to obtain segmentations that are more accurate than any of the individual input segmentations. This paper introduces a new way to combine multiple segmentations using a novel shape-based averaging method. Individual segmentations are combined based on the signed Euclidean distance maps of the labels in each input segmentation. Compared to label voting, the new combination method produces smoother, more regular output segmentations and avoids fragmentation of contiguous structures. Using publicly available segmented human brain MR images (IBSR database), we perform a quantitative comparison between shape-based averaging and label voting by combining random segmentations with controlled error magnitudes and known ground truth. Shape-based averaging generated combined segmentations that were closer to the ground truth than combinations from label voting for all numbers of input segmentations (up to ten). The relative advantage of shape-based averaging over voting was larger for fewer input segmentations, and larger for greater deviations of the input segmentations from the ground truth. We conclude that shape-based averaging improves the accuracy of combined segmentations, in particular when only a few input segmentations are available and when the quality of the input segmentations is low.

Analysis of CT datasets is commonly time consuming because of the required manual interaction. We present a novel and fast automatic initialization algorithm to detect the carotid arteries providing a fully automated approach of the segmentation and centerline detection. First, the volume of interest (VOI) is estimated using a shoulder landmark. The carotid arteries are subsequently detected in axial slices of the VOI by applying a circular Hough transform. To select carotid arteries related signals in the Hough space, a 3-D, direction dependent hierarchical clustering is used. To allow a successful detection for a wide range of vessel diameters, a feedback architecture was introduced. The algorithm was designed and optimized using a training set of 20 patients and subsequently evaluated using 31 test datasets. The detection algorithm, including VOI estimation, correctly detects 88% of the carotid arteries. Even though not all carotid arteries have been correctly detected, the results are very promising.

We propose a novel anatomical labeling algorithm for bronchial branches extracted from CT images. This method utilizes multiple branching models for anatomical labeling. In the actual labeling process, the method selects the best candidate models at each branching point. Also a special labeling procedure is proposed for the right upper lobe. As an application of the automated nomenclature of bronchial branches, we utilized anatomical labeling results for assisting biopsy planning. When a user inputs a target point around suspicious regions on the display of a virtual bronchoscopy (VB) system, the path to the desired position is displayed as a sequence of anatomical names of branches. We applied the proposed method to 25 cases of CT images. The labeling accuracy was about 90%. Also the paths to desired positions were generated by using anatomical names in VB.

Image segmentation algorithms derived from spectral clustering analysis rely on the eigenvectors of the Laplacian of a weighted graph obtained from the image. The NCut criterion was previously used for image segmentation in supervised manner. We derive a new strategy for unsupervised image segmentation. This article describes an initial investigation to determine the suitability of such segmentation techniques for ultrasound images. The extension of the NCut technique to the unsupervised clustering is first described. The novel segmentation algorithm is then performed on simulated ultrasound images. Tests are also performed on abdominal and fetal images with the segmentation results compared to manual segmentation. Comparisons with the classical NCut algorithm are also presented. Finally, segmentation results on other types of medical images are shown.

Curves are often used as anatomical features to match surfaces that represent biological objects, such as the human brain. Automated and semi-automated methods for extracting these curves usually rely on local properties of the surfaces such as the mean surface curvature without considering the global appearance of the curves themselves. These methods may require additional human intervention, and sometimes produce erroneous results. In this paper, we present an algorithm that is based on the fast marching method (FMM) to extract weighted geodesic curves. Instead of directly using the local image properties as a weight function, we use the surface properties, together with the global properties of the curves, to compute a weight function. This weight function is then used by the FMM to extract curves between given points. The general framework can be used to extract curves with different global properties. The resulting curves are guaranteed to be weighted geodesic curves without cusps usually introduced by intermediate points through which the curves are forced to pass. We show some results on both a simulated image and a highly convoluted human brain cortical surface.

This paper presents an algorithm for determining regional cerebral grey matter cortical thickness from magnetic resonance scans. In particular, the modification of a gradient-based edge detector into an iso-grey-level boundary detector for reliably determining the low-contrast grey-white matter interface is described and discussed. The reproducibility of the algorithm over 31 gyral regions is assessed using repeat scans of four subjects, and a technique for correcting the misplacement of the grey-white matter boundary is shown to significantly reduce the systematic error on the reproducibility.

This paper assesses the estimation of kinetic parameters from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). Asymptotic results from likelihood-based nonlinear regression are compared with results derived from the posterior distribution using Bayesian estimation, along with the output from an established software package (MRIW). By using the estimated error from kinetic parameters, it is possible to produce more accurate clinical statistics, such as tumor size, for patients with breast tumors. Further analysis has also shown that Bayesian methods are more accurate and do not suffer from convergence problems, but at a higher computational cost.

High resolution MRI images of the beating heart permit observation of detailed anatomical features and enable quantification of small changes in metrics of cardiac function. To obtain approximately isotropic sampling with an adequate spatial and temporal resolution, these images need to be acquired in multiple breath-holds. They are, therefore, often affected by through-plane discontinuities due to inconsistent breath-hold positions. This paper presents a method to correct for these discontinuities by performing breath-hold-by-breath-hold registration of high resolution 3D data to radial long axis images. The corrected images appear free of discontinuities, and it was found that they could be delineated more reproducibly than uncorrected images. This reduces the sample size required to detect systematic changes in blood pool volume by 57% at end systole and 78% at end diastole.

A 4D image registration method is proposed for consistent estimation of cardiac motion from MR image sequences. Under this 4D registration framework, all 3D cardiac images taken at different time-points are registered simultaneously, and motion estimated is enforced to be spatiotemporally smooth, thereby overcoming potential limitations of some methods that typically estimate cardiac deformation sequentially from one frame to another, instead of treating the entire set of images as a 4D volume. To facilitate our image matching process, an attribute vector is designed for each point in the image to include intensity, boundary and geometric moment invariants (GMIs). Hierarchical registration of two image sequences is achieved by using the most distinctive points for initial registration of two sequences and gradually adding less-distinctive points for refinement of registration. Experimental results on real data demonstrate good performance of the proposed method in registering cardiac images and estimating motions from cardiac image sequences.

A new method for MR to X-ray registration is presented. Based on training data, consisting of registered multispectral MR and CT data, a function is defined that maps multispectral MR data to CT-like data. For new subjects for which multispectral MR data have been acquired, the mapping function is used to generate a corresponding CT-like dataset. The CT-like image is subsequently used for registration to X-ray data, using gradient-based registration. Preliminary experiments indicate that MR to X-ray registration using this method is more accurate and has a larger capture range than gradient-based registration applied directly to MR data.

This paper describes a new piecewise rotational transformation model for capturing the articulation of joints such as the hip and the knee. While a simple piecewise rigid model can be applied, such models suffer from discontinuities at the motion boundary leading to both folding and stretching. Our model avoids both of these problems by constructing a provably continuous transformation along the motion interface. We embed this transformation model within the robust point matching framework and demonstrate its successful application to both synthetic data, and to serial x-ray CT mouse images. In the later case, our model captures the articulation of six joints, namely the left/right hip, the left/right knee and the left/right ankle. In the future such a model could be used to initialize non-rigid registrations of images from different subjects, as well as, be embedded in intensity-based and integrated registration algorithms. It could also be applied to human data in cases where articulated motion is an issue (e.g. image guided prostate radiotherapy, lower extremity CT angiography).

Deforming a digital atlas towards a patient image allows the simultaneous segmentation of several structures. Such an intersubject registration is difficult as the deformations to recover are highly inhomogeneous. A priori information about the local amount of deformation to expect is precious, since it allows to optimally balance the quality of the matching versus the regularity of the deformation. However, intersubject variability makes it hard to heuristically estimate the degree of deformation. Indeed, the sizes and shapes of various structures differ greatly and their relative positions vary in a rather complex manner. In this article, we perform a statistical study of the deformations yielded by the registration of an image database with an anatomical atlas, and we propose methods to re-inject this information into the registration. We show that this provides more accurate segmentations of brain structures.

This paper presents a robust and fully automated registration algorithm for registration of images of Quadrature Tomographic Microscopy (QTM), which is an optical interferometer. The need for registration of such images is to recognize distinguishing features of viable embryos to advance the technique for

In Vitro

Fertilization. QTM images a sample (live embryo) multiple times with different hardware configurations, each in turn producing 4 images taken by 4 CCD cameras simultaneously. Embryo movement is often present between imaging. Our algorithm handles camera calibration of multiple cameras using a variant of ICP, and elimination of embryo movement using a hybrid of feature- and intensity-based methods. The algorithm is tested on 20 live mouse embryos containing various cell numbers between 8 and 26. No failure thus far, and the average alignment error is 0.09 pixels, corresponding to the range of 639 and 675 nanometers.

In inter-subject registration, one often lacks a good model of the transformation variability to choose the optimal regularization. Some works attempt to model the variability in a statistical way, but the re-introduction in a registration algorithm is not easy. In this paper, we interpret the elastic energy as the distance of the Green-St Venant strain tensor to the identity, which reflects the deviation of the local deformation from a rigid transformation. By changing the Euclidean metric for a more suitable Riemannian one, we define a consistent statistical framework to quantify the amount of deformation. In particular, the mean and the covariance matrix of the strain tensor can be consistently and efficiently computed from a population of non-linear transformations. These statistics are then used as parameters in a Mahalanobis distance to measure the statistical deviation from the observed variability, giving a new regularization criterion that we called the statistical Riemannian elasticity. This new criterion is able to handle anisotropic deformations and is inverse-consistent. Preliminary results show that it can be quite easily implemented in a non-rigid registration algorithms.

This paper presents a new approach for lung registration and cardiac motion prediction, based on a 3D geometric model of the left lung. Feature points, describing a shape of this anatomical object, are automatically extracted from acquired tomographic images. The ”goodness-of-fit” measure is assessed at each step in the iterative scheme until spatial alignment between the model and subject’s specific data is achieved. We applied the proposed methods to register the 3D lung surfaces of 5 healthy volunteers of thoracic MRI acquired in different respiratory phases. We also utilized this approach to predict the spatial displacement of the human heart due to respiration. The obtained results demonstrate a promising registration performance.

We present a simple and rapid method for generation of perspective digitally rendered radiographs (DRR) for 2D/3D registration based on splat rendering. Suppression of discretization artefacts by means of computation of Gaussian footprints – which is a considerable computational burden in classical splat rendering – is replaced by stochastic motion of either the voxels in the volume to be rendered, or by simulation of a X-ray tube focal spot of finite size. The result is a simple and fast perspective rendering algorithm using only a small subset of voxels. Our method generates slightly blurred DRRs suitable for registration purposes at framerates of approximately 10 Hz when rendering volume images with a size of 30 MB on a standard PC.

The aim of this project is to verify the accuracy of positron emission tomography (PET) in identifying the tumour boundary and eventually to enable PET-guided resection with removal of significantly smaller margins. We present a novel use of an image-guided surgery system to enable alignment of preoperative PET images to postoperative histology. The oral cancer patients must have a high resolution CT scan as well as undergoing PET imaging. Registration of these images to the patient during surgery is achieved using a device that attaches to the patient’s upper or lower teeth. During the procedure markers are placed around the lesion within tissue that is to be resected. These are marked along with any convenient anatomical landmarks using the image guidance system, providing the location of the points in the preoperative images. After the sample has been resected, slices through at least 3 of these points are made and photographed. Registration should be possible using these landmarks, but the accuracy of alignment is much improved by marking the bone surface in the histology image and registering to preoperative CT.

In this paper we present an enhanced method for non-rigid registration of volumetric multi-modal images using Mutual Information (MI). Based on a hierarchical subdivision scheme, the non-rigid matching problem is decomposed into numerous rigid registrations of sub-images of decreasing size. A thorough investigation revealed limitations of this approach, caused by a peculiar behavior of MI when applied to regions covering only a limited number of image pixels. We examine and explain the loss of MI’s statistical consistency along the hierarchical subdivision. We also propose to use information theoretical measures to identify the problematic regions in order to overcome the MI drawbacks. This does not only improve the accuracy and robustness of the registration, but also can be used as a very efficient stopping criterion for the further subdivision of nodes in the hierarchy, which drastically reduces the computational costs of the entire registration procedure.

Matching 3D shapes is important in many medical imaging applications. We show that a joint clustering and diffeomorphism estimation strategy is capable of simultaneously estimating correspondences and a diffeomorphism between unlabeled 3D point-sets. Correspondence is established between the cluster centers and this is coupled with a simultaneous estimation of a 3D diffeomorphism of space. The number of clusters can be estimated by minimizing the Jensen-Shannon divergence on the registered data. We apply our algorithm to both synthetically warped 3D hippocampal shapes as well as real 3D hippocampal shapes from different subjects.

Electromagnetic tracking systems have the potential to track instruments inside the body because they are not limited by the line of sight constraints that characterize optical tracking systems. To integrate an electromagnetic tracking device into a surgical navigation system, accurate registration is required. We present a two-stage registration mechanism designed to be more accurate than the widely used global fiducial-based registration method. The first stage uses a hybrid Iterative Closest Point (ICP) registration method and the Simulated Annealing (SA) optimization algorithm, to increase the initial registration accuracy. The second stage exploits multiple implanted tracking needles that are used to calculate the affine transform based on the initial transform information, and thereby to compensate for the deformation in real time. Phantom and swine studies have demonstrated the utility of this technique.

A method is presented for the registration of tracked B-mode ultrasound images to a CT volume of a femur or pelvis. This registration can allow tracked surgical instruments to be aligned with the CT image or an associated preoperative plan. Our method requires no manual segmentation of either the ultrasound images or the CT volume. The CT and US images are processed to produce images where the image intensity represents the probability of the presence of a bone edge. These images are then registered together using normalised cross-correlation as a similarity measure. The parameter which represents the speed of sound through tissue has also been included in the registration optimisation process. Experiments have been carried out on six cadaveric femurs and three cadaveric pelves. Registration results were compared with a “gold standard” registration acquired using bone implanted fiducial markers. Results show the registration method to be accurate, on average, to 1.7mm root-mean-square target registration error.

Optimal port placement is a delicate issue in minimally invasive endoscopic surgery, particularly in robotically assisted surgery. A good choice of the instruments’ and endoscope’s ports can avoid time-consuming consecutive new port placement. We present a novel method to intuitively and precisely plan the port placement. The patient is registered to its pre-operative CT by just moving the endoscope around fiducials, which are attached to the patient’s thorax and are visible in its CT. Their 3D positions are automatically reconstructed. Without prior time-consuming segmentation, the pre-operative CT volume is directly rendered with respect to the endoscope or instruments. This enables the simulation of a camera flight through the patient’s interior along the instruments’ axes to easily validate possible ports.