Show Archived Projects

Analysis of Retinal Optical Coherence Tomography
Optical Coherence Tomography (OCT) is a non-invasive, depth-resolved imaging modality that has become a prominent ophthalmic diagnostic technique. We develop segmentation and analysis algorithms for OCT.

Diffusion MRI
The advent of diffusion tensor imaging (DTI) has provided clinicians with the ability to assess the integrity of the brainís neural pathways in a non-invasive manner. This unique ability has quickly allowed DTI to become an established imaging protocol generating images that, like other medical images, benefit from having computational methods for processing and analysis. We work on developing methods for the processing, analysis, and visualization of diffusion MRI images.

Skin Hair Simulator Software
Hair occlusion is one of the main challenges facing automatic lesion segmentation and feature extraction for skin cancer applications. Since hair shafts are very thin with spatially-varying width, validating disocclusion algorithms based on manual preparation of ground truth masks (with accurate hair-width for a large number of hair pixels) would be exorbitantly tedious. In this project, we have developed the first skin hair simulation software.

MIA using ITK on Apple iOS
Motivated by the importance of performing medical image analysis using ITK for many health applications and by the ubiquitous mobile devices, in particular Apple's iOS devices (iPod touch, iPhone, and iPad) we provide the documentation and code to facilitate the development of more sophisticated medical image analysis applications on iOS devices.

Fast Medical Image Analysis
Medical image analysis (MIA) algorithms can be very computationally demanding because of 3D and higher dimensionality of the data as well as the large number of images to be analyzed. This is creating a clear need for speeding-up MIA algorithms. Further, interactive medical image analysis software places even more stringent requirements for close-to-real-time analysis. We are working on developing faster versions of several MIA algorithms.

VascuSynth: Simulation of Branching Tubular Structures for Validation and Learning

Analysis of Tongue Dynamics from US
We are developing techniques for analyzing the movement of the human tongue as captured in 2D dynamic ultrasound.

Reliability-Driven, Spatially-Adaptive Regularization for Medical Image Analysis
Computer vision and medical image analysis problems, e.g. segmentation and registration, are typically formulated as minimization of a cost function. The cost function comprises data fidelity (external energy or likelihood) and regularization (internal energy or prior) terms. The choice of the weight balancing the trade-off between these two terms can have significant effect on the result. Previous methods relied on a training data set for finding optimal weights for a class of images. We propose a spatially adaptive regularization weight that is derived from the image data itself, and relies on estimates of noise, reliability-gated edges, texture, and curvature estimates. We applied our approach, with promising results, to segmentation and registration problems.

Probabilistic Multi-Label Shape Representations
Sources of uncertainty in medical images demand uncertainty-encoding shape representations. We propose the Isometric Log Ratio (ILR) transformation as a probabilistic multi-region representation and show how it enables a multi-region, probabilistic, convex segmentation with shape prior constraints.

MCCAP: Minimal Corpus Callosum Area Plane
We propose an alternative to the mid-sagittal plan (MSP) based on the role of the corpus callosum as a bottleneck structure in determining the rate of inter-hemispheric neural transmission. We designate this plane as the Minimum Corpus Callosum Area Plane (MCCAP).

Perceptual Visualization of High-Dimensional Medical Images
High-dimensional data is becoming more prevalent in general, and in medical imaging applications, in particular. We are developing colour visualization methods that respect the underlying manifold-valued data and applying our methods to various modalities, e.g. dynamic PET / SPECT and Diffusion Tensor MRI.

Fast Random Walker
Updating segmentation results in real-time based on repeated user input is a reliable way to guarantee accuracy, paramount in medical imaging applications, while making efficient use of an expert's time. We developed an algorithm to speed up the popular, globally optimal, probabilistic Random Walker segmentation method, based on offline pre-computation, taking advantage of the time images are stored on servers prior to the analysis session.

Surface And VOlumetric Registration (SAVOR)
Constructing a one to one correspondence between whole brain MR image scans is a problem of critical importance in neuroimaging analyses. We have developed a framework to combine the strength of both surface-based and volumetric-based analyses for consistent, bijective data transfer between brain coordinate systems. By combining a good volumetric registration, with a topology preserving projection from one surface to the other, anatomical surfaces can be registered accurately. SAVOR yields registrations with high correlation of cortical biomarkers and little misregitration of cortical parcellation labels.

To produce accurate segmentations of 2D and 3D structures, manual intervention is often unavoidable. We are developing techniques that requires only minimal, intuitive user interaction for guiding the segmentation.

3D Live-Wire
Accurate and automatic 3D medical image segmentation remains an elusive goal and manual intervention is often unavoidable. We are working on techniques that allow the user to provide minimal intuitive interaction for guiding the 3D segmentation process.

Shape denoising via Template Injection using Binary LDDMM
This project focusses on the generating highly smooth segmentations of manual tracings via Template Injection using binary Large Deformation Diffeomorphic Metric Mapping (LDDMM).

Eye-gaze Driven Interactive Image Segmentation
We develop a hands-free interactive image segmentation using an eye-gaze tracking system.

Motion Correction in Medical Imaging

Groupwise Medial Axis Transform
We augment the traditional medial axis transform with an additional coordinate stored at each medial locus, indicating the confidence that the branch on which that locus lies represents signal and not noise. This confidence is calculated based on the support given to that branch by corresponding branches in other skeletons in the group. This method is used to produce a fuzzy skeleton and to perform intelligent pruning

DeformIt: Simulation of Ground Truth Data
The problem of scarcity of ground-truth, expert annotated medical image data is a serious one that impedes the training and validation of medical image analysis techniques. We develop algorithms for the automatic generation of large databases of annotated images from a single reference dataset and provide a web-based interface through which the users can upload a reference data set and download an arbitrary numbers of novel ground- truth data.

Spine Shape Analysis

SMRFI: Shape Matching via Registration of Feature Images
We perform shape matching by transforming the problem into an image registration task. At each vertex on the shape, we calculate a shape feature and encode this feature as image intensity at appropriate positions in the image domain. Calculating multiple features at each vertex and encoding them into the image domain results in a vector-valued feature image. Establishing point correspondence between two shapes is thereafter treated as a registration problem of two vector-valued feature images. With this shape representation, various existing image registration strategies can now be easily applied. These include the use of a scale-space approach to diffuse the shape features, a coarse-to-fine registration scheme, and various deformable registration algorithms.

MR Neurography of the Sciatic Nerve
This project focuses on the study of the sciatic nerve through MR neurography. It has focussed on the development of MR protocols for imaging uninjured peripheral nerves, and the construction of computational measurement techniques for several key characteristic features of nerves. Work has also been done to create visualization tools based on rapid prototyping technologies.

3D Shape Descriptors for Human Peripheral Nerves
This project is a collaboration with Dr. Andy Hoffer of the SFU School of Kinesiology. We are working on developing shape descriptors for peripheral nerves. One way to describe them are through their skeletons or centerlines. Skeletons are useful representations for nerves as they contain most of the information that one would want like length and the number of bifurcations. Various skeletonization programs were researched and tested to find how good they are when applied to our datasets. A suitable program that uses Voronoi diagrams to get the medial axis was found and it was run using the isosurfaces of the nerve objects.

Shape Matching using Ant Colony Optimization
We have developed the first Ant Colony Optimization algorithm specifically aimed at solving the Quadratic Assignment Problem for establishing shape-correspondence, with proximity information incorporated.

Image Crawlers
Image Crawlers, a new breed of Deformable Organisms, are equiped with 3D tubular medial-based bodies, a new repertoire of sensory modules (e.g. Hessain-based, hemispherical sensors), behavioral routines (e.g. grow, spawn children branch cralwers), and decision making strategies (e.g. branch detection, growth direction). They crawl along tubular and tree-like structures in medical images, segmenting boundaries, detecting and exploring bifurcations, and providing sophisticated, clinically-relevant structural analysis.

Computational Cardiac Anatomy: 3D analysis of heart function
In collaboration with Dr. Elliot McVeigh of the Johns Hopkins University, we are developing algorithms and tools for the quantitative analysis of myocardial function. Towards this, we research on myocardial motion and strain estimation, and population based strain statistics from tagged MRI datasets.

Symmetric Large-Deformation Registration
Medical image registration is the task of finding the topology-preserving transformation between two images, A and B, which brings them into correspondence. One problem with many current methods is that transformation depends on the ordering of the images. We have developed large-deformation registration tools [1] which are symmetric with respect to the images.

Functional Magnetic Resonance Imaging Data Analysis
The goal of this project is increasing the accuracy of fMRI statistical analysis through accurate group normalization. The collaborators are Dr. Lei Wang and Dr. Deanna Barch from Conte-Center, Washington University.

Robust Cortical Thickness Measurement from MRI
This project focuses on the development of robust computational tools for the measurement of cortical thickness. Cortical thickness is a measure of brain shape that has been found to change in some neurodegenerative diseases, including Alzheimer's Disease, AIDS, and Parkinson's Disease. Reliable thickness measurements may lead to techniques for early diagnosis of these diseases, as well as distinguishing between diseases with similar cognitive effects.

Accurate Localization of MEG Functional Data Using Head Shape Registration
We investigated two new techniques in which the we use external features of the head in subject-to-atlas registration, avoiding the acquisition of MR images per subject. The first method involves placing landmarks on a subject's head exterior and performing affine registration, and the second uses a non-linear fluid registration technique (LDDMM) on the external head-shape of a subject.

FreeSurfer-Initiated Putamen, Cadate and Thalamus Segmentation in MRI Using Large Deformation Diffeomorphic Metric Mapping
We describe a new algorithm for the automated segmentation of the caudate (Caud) putamen (Put) and thalamus (Thal) in clinical Magnetic Resonance Imaging (MRI) scans. Large Deformation Diffeomorphic Metric Mapping is performed on Freesurfer-initiated templates to generate segmentation results. MR images of 24 brains (including Parkinson's diseased and Control) are used to test the algorithm. The results are compared with manual segmentations under different measurements of similarity.

Artificial Life Approaches to Medical Image Analysis
We are developing techniques for analysis of medical images based on modeling and utilizing knowledge about the underlying anatomy in the image. We are wokring on developing intelligent deformable models (deformable organisms) that live in the image space and whose goal is to locate and label anatomy.

3D Shape Analysis and Visualization
We are developing novel approaches and tools for the problem of the quantitative and qualitative analysis and visualization of 3D shapes. The aim is to apply these approaches to medical problems of anatomical shape analysis.

Large Deformation Metric Mapping Tools for Non Rigid Registration
We are working on developing new tools for the registration of point, surface, volumetric and DT-fiber orientation datasets. These tools allow the transformation of information into intrinsic coordinates of a template for building of statistical atlas.

n-SIFT for Matching Medical Images
We extended the well known computer vision technique, SIFT, to arbitrary dimensions and applied to matching medical images.

Musculoskeletal Image Analysis
We are developing tools for the analysis of medical imaging data for quantification, visualization, and understanding musculoskeletal anatomy and function and their relation to diseases.

Computational Cardiac Anatomy
We are utilizing canine cardiac DTMRI data to determine the biomechanical properties of the heart. We are developing new techniques for processing, smoothing, and analyzing this data.

Multi-Modal Medical Image Registration
We are working on non-rigidly registering multi-modal images, including nuclear medicine images to x-ray CT using mutual information and intensity correlation based similarity metrics.

Quantifying neuro-degeneration of the Basal Ganglia in Huntington's Disease
The caudate and the putamen are deep nuclei in the brain that are known to undergo atrophy in patients with Huntington\'s Disease. We are working on tools to segment these nuclei from MRI images using LDDMM and quantify the the change in shape and form over time as the disease progresses.

Interactive, Intuitive, and Controlled Shape Deformations
We are developing algorithms for modeling deformable shapes providing intuitive and controlled deformations for use in image segmentation and shape analysis.

Analysis of MR Images for Multiple Sclerosis Studies
We are developing tools and techniques for segmentation and shape analysis of human brains for multiple sclerosis, which will assist in serial studies related to the progression of the pathology.

Analysis and Visualization of Time Varying Medical Image Data
We are developing algorithms for visual exploration and quantitative analysis of spatio temporal medical imaging data such as dynamic SPECT and dynamic PET.

ITK Deformable Organisms Framework
We are developing a new ITK-based Deformable Organisms framework. The framwork facilitaties the design of geometrical, dynamic, behavioral, and cognitive layers, and perception capabilities by making use of ITK classes and coding style.

A wrapper for the ITK toolkit allowing ITK algorithms to be called in MATLAB.

Show Archived Projects

Site by Faculty of Applied Science web team, Simon Fraser University