Knowledge: Mathematics (Spherical Harmonics, Fourier Transform and similar), basic programming skills in C++, Mathematica

The problem of extracting the motion in every frame of an image sequence is usually not solvable using the information present in the image sequence alone. For example, if a single-colored, textureless disc rotates, no movement can be observed. This problem also manifests itself if a rectangle is moving sideways. In that case, no motion can be seen if we only look at a small region around the upper edge of the rectangle (where we can not see the sides). The human visual system analyses the image sequence at different scales, and can therefore correctly estimate the movement at the upper edge of the rectangle. This is possible because it regards the upper edge as a part of the entire rectangle.

The most wide-spread approach in traditional motion estimation, or optic flow, is based on the use of local derivatives of the image sequence to estimate movement. Since only the information of a small region region is used to estimate the flow at a certain position, the above mentioned problem is not solved. An existing method, which uses the motion of so-called toppoints in an image sequence, does not have to cope with this shortcoming, because toppoints exist at different scales in an image. Therefore their motion describes the movement of image structures with different scales, similar to the approach of the human visual system.

We propose an optic flow estimation method that combines these two approaches into one, using the benefits of one to overcome the drawbacks of the other. Perturbation theory is used for the implementation and we perform an evaluation using test results from three image sequences. During this evaluation a problem in the motion estimation is encountered that leads to a great inaccuracy. Closer inspection reveals that this problem is caused by an erroneous relation between toppoint motion and the flow field, on which the existing toppoint method is based. Some alternatives to this erroneous relation are presented, which unfortunately do not appear to be the correct solution. It is shown, however, that if this problem can be overcome, our approach is expected to provide performance comparable to other state-of-the-art optic flow techniques.

For more information please contact Bram Platel

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is crucial in the work up of findings in the breast that cannot be assessed from X-ray or ultrasound images, and in the screening of the high-risk population. The high-risk population consists of women that have a personal or familial history of breast cancer, or that are known to be carriers of certain gene mutations (for instance BRCA1 and BRCA2). DCE-MRI is a very sensitive technique (most tumors are detected), but the specificity is low (too many lesions are suspect of malignancy, resulting in unnecessary biopsies). Computer-aided diagnosis could be a powerful tool to improve the specificity of DCE-MRI of the breast.

 

During a DCE-MRI breast exam, a low-molecular-weight contrast agent, commonly gadolinium-based, is injected. The contrast agent causes enhancement in the tissue by shortening the longitudinal relaxation time (T1). The uptake of contrast agent is therefore measured with a series of T1-weighted images over time. Due to the process of tumor angiogenesis, contrast-agent uptake is often more rapid and higher in tumor tissue compared to normal tissue. The DCE-MR image series serves to assess the tissue on a morphological basis (spatial patterns) and kinetic basis (temporal/uptake patterns); the latter being the topic of this master project.

 

Kinetic analysis, as it is currently performed in clinical practice, is based on descriptive curve-shape analysis. This means that curve properties like time-to-peak and maximum enhancement are calculated. Newer analysis methods are however aiming to describe curve-shapes in terms of the physiological processes underlying the observed uptake of contrast agent. Physiology-based modeling is gaining attention, but it still has to prove its added value in the field of clinical diagnostics.

 

One of the issues in the field of clinical diagnostics is that commonly the temporal resolution of the T1-weighted dynamic image series is low; whereas physiology-based modeling requires data with high temporal resolution. Due to this mismatch, it is hard to assess the usefulness of physiology-based modeling. However, without demonstrating the possible advantages of acquiring data at higher temporal resolution, the clinical practice will not change. The purpose of this project is therefore to explore the possible impact of physiology-based modeling on diagnostics: does physiology-based kinetic modeling of high-temporal resolution data lead to useful kinetic features? To enable this research, several data sets (~ 20) were acquired at high temporal resolution.

 

Volume representations of blood vessels acquired by 3D rotational angiography are very suitable for diagnosing a stenosis or an aneurysm. For optimal treatment, the shape parameters of the diseased vessel parts are needed by physicians. Therefore, a fully-automatic extraction of this shape from such a volume representation has been developed. The demo program v3d_main has been developed to test the various algorithms, such as the segmentation algorithm, the wave propagation algorithm and the thinning algorithm.
This paper first discusses and analyses the blood branch labeling acceleration algorithms, and proposes two methods for improvement. The first one is called the surface wave propagation method which restricts the wave moving only along the blood vessel surface. This method is applied to detect the extremities of the vessel voxel structures. The second one combines a thinning algorithm with the surface propagation to extract the center lines and the bifurcations of the blood vessels, which also gives the vessel voxels a unique number (label) per vessel branch.
Proper validation results are given in this report. The result shows that the two methods can substantially decrease the computation time and keep the labeling accuracy. However, whether the branch labeling result, generated by surface wave propagation based on a vessel graph, are suitable for computer assisted diagnosis has not been investigated yet.

This thesis provides an automatic method for the segmentation of the hippocampus region and the detection of Alzheimer’s disease based on the size of the regions. The hippocampus is known to shrink in time due to cell death, and it is linked with increased memory loss, which is the primary symptom of Alzheimer’s disease. First, the hippocampus region has been analyzed by a region-of-interest mask defined for the coronal views and was determined according to the Medial Temporal lobe Atrophy (MTA). To compare this template-based method, we do also perform the segmentation of the hippocampus based on several low level segmentation methods, such as, region growing, edge-based segmentation and active contours. However, none of these methods achieves the desired result because of the complex appearing of regions surrounding the hippocampus and the low image contrast in the hippocampus region. As a result, we propose to use an active shape model to represent the shape of hippocampus based on eigenvectors extracted from training shapes. This model allows for more freedom of shape changes, such as, the affine transformation. Mutual information is used to evaluate the segmentation label. After full segmentation of the hippocampus, more information, such as, pixels of CSF surrounding the hippocampus could be extracted using local region growing.
With respect to sensitivity, specificity and ROC curve, features extracted from the segmentation of the hippocampus region lead to the best result. Moreover besides features extracted from baseline data, CSF changes within a period (roughly one year) is also helpful for computer diagnosis.

Deep Brain Stimulation is the implantation of electrodes in the subthalamic nucleus (STN), a brain area that displays strong changes in Parkinson’s Disease patients.
Stimulation results in significant improvements of the patient’s locomotion. However, in half of the cases, changes also occur in the behavior or intellectual capabilities of the operated patient, such as a depression or mania. It is assumed that the side-effects will be less if you insert the electrode in a more accurate way, meaning only in the motor part of the STN.

To achieve this goal, it is important to localize the STN and its motor part. We think that data acquired by anatomical MRI, diffusion-weighted MRI and functional MRI, can help to visualize the STN and its motor part.
In September and October 2008, we have acquired anatomical, diffusion-weighted and functional data of 12 healthy volunteers. The goal of this project is to find features in anatomical, diffusion-weighted and functional MRI images that can be combined to find the subthalamic nucleus.

This Master Thesis project, performed at the Biomedical Image Analysis group (Eindhoven University of Technology, department of Biomedical Engineering), is focused on the practical application of orientation scores using dedicated hardware resources. We will show the potential of orientation scores in real time quantification of collagen fibers in 2D slices taken from the fibrous tissues around the bone under strong hardware restrictions. The quantification is needed for mechanical studies where some of the relevant features are the orientation of the collagen fibers, the coherence of the fibrous structure and the curliness of the fibers. With coherence is meant in this context the presence of local organized patterns in the tissue like groups of aligned fibers, the pattern that we find in the periosteum around the bone, or the orthogonal organizations that appear on cardiac tissues.

The aim of this project is to implement a white matter tractography method based on
anisotropic wavefront propagation in diffusion tensor images. The next task is to
assess and analyze the results of different connectivity measurements in order to
further develop the DTItool. Once we have selected a measurement that can give
more coherent results, we intend to apply the methodology described above to fiber
tracking in rat or human brain diffusion tensor images.
We are looking for ways to improve deep brain stimulation procedures. Our main
interest is a more specific targeting of the motor part of the subthalamic nucleus,
stimulation of which enhances motor function in Parkinson’s Disease patients.
We assume that an investigation of the connectivity of the subhalamic nucleus with
other brain structures like the motor cortex, globus pallidus, and substantia nigra can
help to achieve this goal. Because the tracts of interest are smaller and probably more entangled than major bundles like the corpus callosum and the pyramidal tracts, we need sophisticated tractography techniques.

Predicting functional human movement becomes more and more important. Often, minimal invasive surgery techniques are used to maintain as much functionality of the operated tissues as possible.The patient-specific questions and the anticipated options or strategies during surgery may be simulated on beforehand with Computer Assisted Surgical Planning (CASP) software. Only when this CASP-software is based on functional, patient-specific musculoskeletal models will it be possible to predict the functional outcome of surgery. DTI (diffusion tensor imaging) has proven to be able to determine muscle geometry.

The aim of this project is to extend the DTI-Tool (developed at the BMIA group) with a fiber segmentation algorithm based on a minimal distance of a fiber end (DTI) to a tendon plate (T1-W)

Atrial Fibrillation (AF) is a disorder characterized by quivering atria, beating inefficiently and preventing them to be completely emptied. AF puts people at risk of blood clots and stroke. By means of ablation, i.e. surgically removing faulty electrical path ways in the heart by applying heat to the conductive tissue, patients with AF can be helped.
During ablation interventions, image guidance is mostly obtained from fluoroscopy. Navigation by fluoroscopy is difficult, and therefore alternative methods for guidance, e.g., by means of augmented reality will be investigated. Such augmented reality requires presegmentation of, e.g., the Left Atrium (LA) and subsequent mapping of the segmentation result onto the live fluoroscopy imagery during the intervention.

Cerebral Visual Impairment (CVI) is a neurological disorder caused by brain damage to the optical path located posterior to the optic chiasm. Due to this brain damage patients have deficits in aspects of visual perception and processing of visual information. Depending on the affected cortical areas this may lead to poor object and face recognition, reduced perception of depth and orientation problems.
At present, the diagnosis of CVI is based on neurological and psychological examinations and observations. These tests are time consuming, subjective and require cooperation of an impaired child. Because of this, a new objective diagnostic tool is needed.

At the department of Neuroscience, Erasmus MC, Rotterdam, a new diagnostic tool is proposed. This new tool is based on the fact that eye and/or head responses are automatically induced towards visual feature when it is detected, so called preferential looking. A range of visual stimuli were displayed on a monitor with an automated eye-tracking system for objective gaze measurements.
Each trail contains 30 attractors to attrack the attention of the subjects, 4 motion and form coherence sequences to measure the global processing of the dorsal and ventral streams, 16 competitive and non competitive dots to exam an inhibitory effect of competitive tasks, 4 dot saccades to measure visual perception and latencies and 4 smooth pursuits to test the smooth pursuit system.
We tested 120 children (0-10 years) without and 13 children (2-4 years) with cerebral pathology. Based on the visual stimuli applied to both groups, 2 new diagnostic markers were defined. Latency was defined as the time between the onset of a visual stimulus projection, e.g. attractor, competive dot, and the first reflexive eye movement towards the quadrant in which this stimulus was displayed. In addition, gaze fixation was tested as diagnostic marker. Gaze fixation area was equal to a fitted ellipse area containing the gaze data in the correct quadrant.

Impaired children showed significantly delayed latency time, larger fixation area and shorter looking time to the form coherence compared to healthy children (p

Automated preferential looking can be applied in both healthy and disabled children. Latency and fixation area can be promising diagnostic markers to asses function loss in children with CVI.

A recent technological advancement in radiotherapy that providesthe possibility to increase tumor control and reduce side effects simultaneously is Intensity Modulated Radiation Therapy (IMRT).IMRT not only uses different beam positions and shapes, but alsovaries the field intensity across the beam (a beam then consistsof segments), in that way the dose is conformed to the tumor.

The possibilities for delivery errors increase, because the therapybecomes more complicated. This makes pre-treatmentverification, to ensure that the delivered dose agrees with thecalculated dose, very important. A quick, accurate and reliablemethod for this pre-treatment verification will be developed.

Prostate cancer is one of the most common cancers in men and a major cause of death. In the Netherlands, 8.000 new cases were diagnosed in 2004 and 2.400 men died that year of prostate cancer. Because prostate cancer grows slowly, it can, generally speaking, be effectively treated when it is detected at an early stage. However, physical problems usually related to urination can occur when the cancer has had an opportunity to grow. From the new diagnosed cases, 75% of the men in question were 65 years or older.

In diagnoses and treatment, the determination of lymph node involvement is an important issue. The lymph nodes in the pelvic region give information about the spreading of the cancerous cells. When cancerous cells travel through the lymphatic system the lymph nodes in the pelvic system are the first to be infected. Presently, imaging techniques are becoming more important in the detection of metastases in lymph nodes. Especially the usage of different MRI techniques in combination with new contrast agents based on ultra small paramagnetic iron oxide particles (USPIO-particles, Sinerem) have improved the sensitivity and specificity of the detection of lymph node metastases.

Detection of lymph nodes metastases on MR images is a tedious and time consuming process. Even for an experienced radiologist it can take up to 45 minutes per patient. Development of a computer algorithm that helps the radiologist in finding these lymph nodes would increase productivity and may even help to improve diagnosis. In this thesis such a CAD-algorithm is proposed.

In this report three segmentation techniques have been implemented to segment visible mouse data/magnetic resonance images of a mouse.
The Magnetic Resonance Lab (MRL) in Eindhoven initiated a project to scan a mouse and its organs. By visualizing the organs of the mouse on the computer a 3D database of the mouse is created. This database can serve as a reference for researchers that are performing experiments on mice. Before visualization the organs must be segmented.

With the introduction of IMRT, the position and reproducibility of the leaves become more important, especially for abutting segments. A calibration error is a systematic error in leaf positioning.
The goals are:
• the development of a leaf verification method applicable to a Siemens linac using a CCD-camera based EPID.
• the verification of the leaf positions of two different types of Siemens MLCs and to check whether the MLCs and the calibration method using the light field are accurate enough for IMRT.

Megavoltage Cone Beam CT (MV CBCT) can be used for 3D imaging of the patient anatomy on the treatment table. To use MV CBCT images for dose calculation purposes, reliable electron density (ED) distributions are needed. Patient scatter, beam hardening and softening effects result in cupping artifacts in MV CBCT images.
A method based on transmission images is presented to correct for these effects without using prior knowledge of the object’s geometry.

Purpose: To investigate the feasibility of a qualitative automated error detection method based on parameters from a gamma evaluation which can be used to evaluate deviations in portal dose distributions.

Methods and materials: The accuracy of the evaluation method is investigated by performing a simulation study. Portal dose images (PDIs) are generated and perturbations due to machine output variation, set-up errors and organ motion are imposed to the images; a gamma evaluation is applied. In nearly homogeneous regions of a PDI, dose differences are calculated; in regions with dose variations, displacements are determined. A perturbed image is corrected for machine output variations before displacements are determined. Because a portal dose distribution is not invariant to set-up errors and organ motion, portal dose images are also simulated using a twodimensional (2-D) portal dose prediction model based on pencil beam scatter kernels. Using this model, changes in PDIs due to patient perturbations could be simulated. Finally, the error qualification method is tested on clinical PDIs.

Results: A distinction can be made between errors caused by dose differences and displacement of patient or organs. Changes in portal dose distribution caused by patient or organ displacements do not influence the sensitivity of the error qualification method. Accuracy is dependent on the treatment site.

Conclusion: The method that is developed can be useful in clinical practice for qualitative automated error detection using electronic portal dosimetry.

Dynamic PET can be used to obtain time activity curves (TACs) describing the activity profile of a tracer for each independent voxel in time. The uptake kinetics of the tracer is dependent on tracer specific properties as also on the biologic characteristics of underlying tissue. In this study, it was hypothesized that by analyzing TACs it is possible to differentiate the tumor from surrounding tissues better than from static PET measurements or by manual tumor delineation.

Electronic Portal Imaging Devices (EPIDs) are not only suited for patient set-up verification and detection of organ motion but can also be used for dosimetric verification of (complex) treatment techniques or in-vivodosimetry. The aim of this work was to investigate the dosimetric properties of a new commercially available amorphous silicon (aSi) EPID and the feasibility of a calibration procedure to obtain relative full-scatter portal dose distributions.

Maxima Medical Center is specialized in perinatal and neonatal care. In the departments of Gynaecology and Obstetrics of this hospital there are various research projects to improve medical care.

In these departments a new ultrasound modality is used: Kretz Voluson 730, which gives a possibility to perform 3D/4D ultrasound investigations. With 3D-ultrasound all three projections of the
examined fetus together with the volume representation can be displayed. It is also possible to image the movements of a fetus in time (4D ultrasound; x,y,z,t).

With 3D ultrasound the volume (weight) of an fetus can be determined. Volume (weight) is an important measurement to judge the development and condition of the fetus. On the Kretz the volume of the fetus can be determined manually. This operation is complicated and time-consuming. The measurement error of this volume calculation operation is unknown.

The aim of this study is to develop an automatic volume calculation of a fetus based on 3D ultrasound
volume scans.

Around 17.500 people in the U.S. die from primary nervous-system tumors each year. One of the most common treatments for brain tumors is resection. This surgical procedure requires very careful planning and execution. To assist the neurosurgeon with the planning of such delicate procedures a visualization tool should be designed that integrates datasets from various modalities, such as MR, CT, fMRI and DTI (diffusion tensor images) and displays them in an intuitive way. An important requirement is that the design of the tool should be flexible enough to integrate navigational tool tracking information when used for surgical guidance in the operating room itself.

Illustrations are able to transmit and concentrate the attention of the reader (user) to what is important and not to details that are not necessarily relevant. In the last years, inspired by these illustrations the so called illustrative volume rendering has emerged.

 

The goal of this project is to extent Stef Busking work (see finish projects in illustrative Rendering) by implementing a hardware based VolumeFlies framework, using the flexibility of nowadays Graphics Cards. GPU based particle systems for other purposes, like flow visualization, already exist. It is also of interest to improve the specific implementations of the VolumeFlies framework modules.

Dynamic-contrast enhanced MR imaging has become a valuable tool for detection, diagnosis and management of breast cancer, exposing cancerous tissue as areas of mass- or non-mass-like enhancement. Non-mass-like enhancements are often related to early forms of breast cancer and are therewith diagnostic criteria of particular importance. Unlike mass-like enhancements describing compact regions of suspiciously enhancing tissue with distinct morphologic and dynamic characteristics, non-mass-like enhancements are complex distribution patterns of suspiciously enhancing tissue interspersed by areas of normal tissue.

The aim of this project is to analyze the clinical requirements and to develop, implement and evaluate methods that help radiologists to detect, delineate and/or assess areas of non-mass-like enhancements.

Candidates should be interested in mathematics, image processing and/or pattern recognition, but also in the biomedical aspects of breast cancer diagnosis. Experiences in programming with C/C++ are of advantage, but not a prerequisite.

The project is part of a larger collaboration between the TU/e and Philips Medical Systems, Best in the field of computer-aided diagnosis in breast MRI.

In biomedical image analysis one often needs to detect elongated structures like blood vessels, contours, or catheters. Noise and occlusions may interrupt elongated structures. Earlier work focussed on filling gaps in contours using the framework of orientation scores and G-convolution.
The aim of this project is to extent this work by making the G-convolution adaptive with respect to local estimates of the curvature.

In hospitals, image guided surgery is done on data made before an operation. During neurosurgery, the shape of a patient’s brain can change due to brain shift or tumor resection. Because of these tissue deformations, navigation and planning on real time MR data instead of pre-operative data can be of great help during neurosurgery.
The intra-operative Polestar N20 MRI scanner (0.15 Tesla) allows the surgeon to navigate on real-time data using the Medtronic Stealth Station and the Polestar N20. The advantages of the Polestar are the low cost and there is no need for an expensive change of instruments and operating room. The compromise here is the image quality, the low field results in a low resolution and low SNR.
The aim of this project is to register in 3D the pre-operative data and the intra-operative data made by the Polestar N20 to make high resolution information available for navigational purposes during surgery.

The University Hospital Maastricht AZM has acquired an open MRI scanner (Medtronic/Odin Polestar N20) for neurosurgery applications. This is the first low-field open MRI scanner in the Netherlands. The TU/e and azM are scheduling a series of collaborations, in order to improve the use and navigation issues in clinical practice (resection surgery and deep brain stimulation).
The low magnetic field of the polestar makes it usable in the operation room but it inherently involves a lower signal to noise ratio. Surgical planning and inter operative navigation could be improved if the more detailed images from the high field MRI scanners could be warped onto the images from the polestar.

The registration method should be able to deal with the low signal to noise ratio in the data from the polestar. Field inhomogeneities cause deformations in the polestar images. The deformations in the center of the image appeared to be quite small, but increase towards the edges. This has to be taken into account during registration./div>

In the western world, vascular disorders form a major medical problem. To increase the knowledge of the underlying mechanisms of, for example, atherosclerosis, extensive research is performed, to find causes of these disorders. At the university of Maastricht, a special type of microscopy is used, called two photon laser scanning microscopy (TPLSM). Using the TPLSM, three dimensional images can be extracted from viable arteries.
To describe the processes occurring in the large arteries as a reaction on changing circumstances, two methods are proposed. The first method focuses on the estimation of the radius of imaged arteries. The second method can be used for the counting of cells within the vessel wall.
The estimation of the radius of an artery is complicated by the fact that it is often only possible to image a small part of the vessel wall. To get reliable results, a new method has been devised, based on the Hough transform. In this study, it is shown that the proposed method is more accurate than other methods proposed in literature. Using the proposed method, an accuracy expressed in the standard deviation of the error, of 2.5 % can be achieved, against 10 % that is obtained using the often used least squares method.
The second subject that is covered by this work, is the counting of cells. A modelbased
approach has been taken, to fit ellipsoids on sets of potential edge points. Despite of the need for further research, the preliminary results show to be promising. Virtually all nuclei are detected, while after merging over-segmented nuclei, few errors can be identified.

Active Shape Models (ASMs) are statistical shape models extended with a matching component. Thus, by matching an ASM to a (medical) dataset, image segmentation can be achieved.

Shapes are usually described as a set of points obtained by sampling objects. Therefore, ASMs require a point-based update, i.e. an update for each point present in the shape description, to find a new instance of the ASM that matches the data (better). It can happen that at certain positions, such updates cannot be obtained from the data, but still a segmentation has to be obtained. This requires modifications to the standard ASM, to enable matching of an ASM based on a densely sampled shape class to a sparse(r) representation of the shape class in an unseen subject.

This project aims at investigating three different methods for doing so, and compares the accuracy of the segmentation results obtained by the different methods.

MRI of the breast makes use of the fact that tumors show angiogenesis. This can be visualized by using a dynamic, contrast-enhanced T1-weighted scan. The rate of uptake and wash-out of contrast medium is an important sign of malignancy. However, some lesions (most notable early stages of cancer) cannot be identified by the speed of contrast uptake and wash-out alone. Presently, the shape of the enhancing area identifies them to radiologists. A T2 weighted scan is also part of a breast MRI examination. The usefulness of this scan in the classification of malignant and benign lesions is not fully clear. There are indications that these images can improve in classification of breast lesions.

To alleviate tremors, akinesia, and rigidity in Parkinson's Disease patients, chronical deep brain stimulation can be performed. Subthalamic nucleus (STN) is one of the common targets. This nucleus lies just above the brainstem, in the basal ganglia. Navigation towards this core is not without risks, and therefore requires difficult and time-consuming pre-operative planning. This trajectory needs to meet several strict requirements. We propose to develop an automatic trajectory finding algorithm which meets all the requirements of the trajectory. Afterwards, the resulting trajectory should be visualized in 3D, together with the segmented structures, to enable the neurosurgeon to examine the trajectory and correct it if necessary.

The cardiac muscle (myocardium) contains a numerous amount (i.e. millions) of bloodvessels that range in diameter from a few micrometers to several millimeters. To quantify parameters describing the microvessel tree, like bifurcation level, distance between bifurcations, vessel diameter, a.o., first the complete vascular tree has to be extracted from a dataset.

For a proper resolution, a digital camera with an in-plane resolution of approximately 40 micrometers and a cryomicrotome with the ability to cut slices of similar thickness are used to image goat hearts (ex-vivo) in 3D.

Artifacts ocurring due to an assymetric point distribution function are repaired first. Multiscale image analysis is developed and used to extract the blood vessels from the data sets, which are typically in the order of 2000x2000x2000 voxels. Finally, a skeleton from the vascular tree is built from which information describing the tree can be obtained, and which is further used as basis to find the tiny vessels in a more accurate sense, to quantify, e.g., vessel diameter.

Volume rendering is a well established technique that is used for the visualization of medical volume data. Most of the techniques are based in the generation of images based on an approximation of a realistic physical model. However, if we look at the anatomic books, they still use manual illustration instead of photographs or other more realistic means. This project wants to explore illustration techniques for volume rendering and combine them with the more common volume rendering techniques to achieve enhanced images.

MR is becoming an important modality in the diagnosis of breast cancer. The main problem here is not the sensitivity but the specificity of contrast-enhanced MRI of the breast: it will enhance most lesions but also a lot of other tissue. State of the art is to make a dynamic contrast-enhanced T1 weighted scan with 4-6 stacks of 30+ images at time intervals of 90-120 seconds.
As faster scanning protocols become available, it becomes possible to apply more advanced analysis to the contrast uptake of various lesions. In this project, we want to investigate this.

X-ray grids are an indispensable component in any diagnostic X-ray system. This study aims at an increase of the resolution of the digital detector, especially with super-resolution methods. These techniques use sub-pixel shifts of the detector and combine multiple shots of the same object in one better image.

Analysis of 3D 2-photon microscopy images of remodeling collagen fibres in artifical heartvalve tissue.

In biomedical image analysis one often needs to detect elongated structures like blood vessels, contours, or catheters. Noise and occlusions may interrupt longated structures . Thus, standard line detection algorithms may just capture parts of an elongated structure rather than the complete line or contour.
The aim of this project is the development of an algorithm that detects elongated segments in orientation scores and fills the gaps in between with stochastic completion fields suited to the medical task at hand.

MR diffusion tensor imaging (DTI) measures the diffusion of water molecules in tissue. The diffusion is expressed by a second-order tensor. This tensor is an indicator of the underlying structure, for example, the fibers in the brain.
The task of this project will be to transfer the existing DTI tool to the Virtual Reality set-up available at the BIOMIM image-analysis lab.

To make screening of the colon possible with CT, the dose has to be reduced substantially. The goal of this project is to reduce the noise in low-dose CT and to evaluate the quality of the segmentation, visualization and detection.

Two-photon fluorescence microscopy is a relatively new imaging modality. The formation of new tissue during the formation of atherosclerotic plaque of mice is investigated. Goal of the project is to set up an image-analysis environment. Focus will be on shape, motion and textures analysis and 3D visualization of cell structures.

Two-photon fluorescence microscopy is a relatively new imaging modality. The formation of new tissue during the formation of atherosclerotic plaque of mice is investigated. Goal of the project is to set up an image-analysis environment. Focus will be on shape, motion and textures analysis and 3D visualization of cell structures.

Diffusion Tensor Imaging (DTI) is a Magnetic Resonance Imaging (MRI) technique for measuring diffusion in biological tissue. DTI data is difficult to visualize because of the high amount of information available in each sample point. A prominent DTI visualization technique is fiber tracking. The fiber tracking algorithm creates streamlines (fibers) that correspond to the major white matter fiber bundles in the brain. Individual structures are virtually indistinguishable and it is very difficult to extract any useful information. To overcome this problem, we use a clustering algorithm to organize the fibers into groups that are meaningful and anatomically correct.

Design and implementation of a dedicated tool for segmentation of cardiac structures from MR data using Active Contours and knowledge about anatomy and movement of cardiac structures.

Computer aided detection of pulmonary emboli require automatic and appropriate segmentation of the pulmonary vessel-tree. In this master project, methods will be developed to detect the major pulmonary vessels. In a next step, arteries will be separated from veins.

Optic flow describes the apparent motion that is present in an image sequence. We show the feasibility of obtaining optic flow from dynamic properties of a sparse set of so called anchor points. The advantage of approaching the optic flow estimation problem using these anchor points is that in these points the notorious aperture problem does not manifest itself. The proposed optic flow estimation method succeeds in finding a dense vector field that approaches the optic flow field from a sparse set of inherent multi scale anchor points. As opposed to classical optic flow estimation schemes the proposed method accounts for an explicit scale component of the vector field, which could encode a hitherto unknown dynamic property.

Cardiac Electrophysiology procedures are performed under continuous X-ray fluoroscopy surveillance. The goal of the project is to develop image analysis techniques to detect Electrophysiology catheters in the fluoroscopic images. The idea is to improve current methods to detect thin lines (like the catheters) by taking a larger spatial context into account, using the assumptions that catheter-lines are continuous and exhibit finite curvature.

A diffusion tensor image (DTI) is a tensor valued magnetic resonance image that captures the apparent local diffusivity of water molecules in each spatial direction. In this project the goal is twofold, (i) to devise a mathematically well-founded regularization scheme, and (ii) to define elongated structures.

Atherosclerosis is the formation of plaque in the larger bloodvessels. It can lead to serious obstruction of the bloodflow, and rupture may give rise to infarcts in the brain. Understanding the nature of the constituents is essential for diagnosis and treatment planning.
The goal of this project is to automatically find the classification of the different tissues in and around the plaque, by means of non-invasive MRI techniques.

In this project, we will investigate various possibilities for extending the image processing functionality for CT lung diagnosis: lung detection, lung segmentation, bronchia tracking and vessel tracking.

The Visible Mouse is a project of the MR Laboratory to acquire high resolution datasets of mice by means of high-field MRI. The mouse is the new laboratory for molecular imaging and experimental radiology.
In this project a toolkit is developed in Mathematica, based on deformable contours (2D) and surfaces (3D), to segment the 3D datasets of the Visible Mouse. The high level flexible toolkit can incorporate physical models and statistics into the equations for snakes and level sets.

MR diffusion tensor imaging measures the diffusion of water molecules in tissue. The diffusion is expressed by a tensor. This tensor is an indicator of the underlying structure, for example the fibers in the brain. This projects consist in investigating 3D visualization techniques that better allow the interpretation of this tensor data.

For many image processing tasks, hierarchical and topological methods are re-quired. New approaches to image analysis involving the deep structure of images are promising, but many questions about points in scale space are still unanswered. In this thesis, a second order reconstruction algorithm for multi-scale points is presented in order to extract information about these points. It is tested with random points and spatially equidistant points in scale space. Us-ing the equidistant points, an optimal distance between reconstruction points is measured, taking the limited machine precision into account. For this measurement, the condition number of the correlation matrix is used. The algorithm is also tested with multi-scale critical points and multi-scale top points. Two possible applications for the reconstruction from multi-scale top points are discussed: data compression for images using reconstructions and content based image retrieval using coefficients of the reconstruction algorithm. The results of both feasibility studies are promising.

For many image processing tasks, hierarchical and topological methods are required. New approaches to image analysis involving the deep structure of images are promising, but many questions about points in scale space are still unanswered. In this thesis, a second order reconstruction algorithm for multiscale points is presented in order to extract information about these points. It is tested with random points and spatially equidistant points in scale space. Using the equidistant points, an optimal distance between reconstruction points is measured, taking the limited machine precision into account. For this measurement, the condition number of the correlation matrix is used. The algorithm is also tested with multiscale critical points and multiscale top points. Two possible applications for the reconstruction from multiscale top points are discussed: data compression for images using reconstructions and content based image retrieval using coefficients of the reconstruction algorithm. The results of both feasibility studies are promising.

One main task from the MR tagging application is to extract information about deformation of tissue. This information can be described in terms of the optic flow field. The optic flow field (or some people call it the optical flow field), is a vector field that describes an apparent movement of pixels in an image sequence. Our optic flow model is based on the biological visual system, which uses the Gaussian scale-space representation.