[Presentation file for my medical research]
TOPIC-1: CARDIAC DELAYED ENHANCEMENT MAGNETIC RESONANCE IMAGE PROCESSING
MOTIVATION: The coronary arteries supply oxygen and nutrients to the heart muscle. A heart attack (myocardial infarction) occurs when muscle dies or is permanently damaged due to inadequate supply from the coronary vessels. Delayed Enhancement Magnetic Resonance Imaging (DE-MRI) could be used to identify affected tissue and quantify the impact on heart function.
GOAL: Quantification and analysis of left ventricle via segmentation of myocardial borders and the classification of myocardial tissue as viable (healthy) or non-viable (dead).
TECHNIQUES: Active contour models, support vector machines and variational non-rigid registration.
SOME RESULTS:
Endocardium is represented with red contours, epicardium is represented with green contours and non-viable tissue is shown with the pink transparent layer.
RESEARCH STATUS: Main phases of the research completed successfully.
TOPIC-2: CARDIAC TRACKING AT CINE-DEMR SEQUENCES
MOTIVATION: Cine-Delayed Enhancement Magnetic Resonance imaging (Cine-DEMR) is a novel imaging technique targeted to the left ventricle of the heart which combines the advantages of both Cine MR and Delayed Enhancement MR (DEMR). Like Cine imaging, Cine-DEMR recovers the motion of the heart over the cardiac cycle - the detection of motion abnormalities such as hypokinesis in Cine is a well established indicator of cardiac health. Successful (semi)automatic quantification procedure working on CINE-DEMR images could greatly decrease the doctors' analysis time.
GOAL: Tracking of left ventricle providing segmentation of myocardium and the following analysis of segmented tissue.
TECHNIQUES: Active shape models and support vector machines.
SOME RESULTS:
Segmented sequence Cine-DEMR sequence. Endocardium is represented with red contours, epicardium is represented with green contours.
An example tracking/classification result. The pink shaded pixels represent those found to be non-viable. Right: The ROC curve describing the results of our experiments. 96.80% of the area is under the curve.
RESEARCH STATUS: Main phases of the research completed successfully.
TOPIC-3: VESSEL LABELING
MOTIVATION: With the development of better vessel tracking techniques, new and higher level problems like the labeling of these segmented vessels have raised. Successful labeling would be useful in many aspects like providing a feedback for the tracking process (for refining the segmentation results) and giving a complete landmark set that would be used for the (heart) atlas preparation and registration purposes.
GOAL: Labeling of the four main coronary vessels, RCA (right coronary artery), LM (left main coronary artery), CFX (circumflex coronary artery) and LAD (left anterior descending artery).
TECHNIQUES: Advanced ray casting from horizontal, vertical and long axes of the heart.
SOME RESULTS:
Some labeling results: red represents the RCA, green LAD, blue LM and yellow CFX
RESEARCH STATUS: Generalization of the developed techniques to a wide variety of vessels is an ongoing research.
[Report]
TOPIC-4: SPECT - CTA REGISTRATION
MOTIVATION: There exist many techniques for registering cardiac Computed Tomography Angiography (CTA) data to Single Photon Emission Computed Tomography (SPECT). However, there are not many approaches that bring robustness and high speed together. Focusing on the information that can be (always) extracted from the CTA and SPECT data could be useful for having stable landmark sets.
GOAL: Development of a robust and fast registration technique for cardiac CTA and SPECT data.
TECHNIQUES: Random walker, ICP (iterative closest point).
SOME RESULTS:
CTA left ventricle (LV) is fitted to SPECT data with the found deformation field: fitted LV is represented with color
RESEARCH STATUS: Main phases of the research completed successfully.
[Report]
TOPIC-5: CARDIAC DELAYED ENHANCEMENT MAGNETIC RESONANCE IMAGE CLASSIFICATION
MOTIVATION: Quantification of Delayed Enhancement Magnetic Resonance (DEMR) cardiac images is a binary classification problem; a tissue is either viable or non-viable. Existing approaches try to solve this classification problem via processing intensity histogram in various ways. However, non-viable tissues mostly define closed regions that let experts, who do the analysis manually, to do the classification by not only using the intensity but also the shape priors. Defining a model favoring specific shape (closed) and intensity values should be a good solution for this specific problem.
GOAL: Classification of left ventricle myocardium tissue as viable or non-viable.
TECHNIQUES: Binary classification with graph-cuts.
SOME RESULTS:
Bottom-left is the original image, bottom-right is the myocardium mask, up-left is the classification result on the original image, up-right is the classification.
RESEARCH STATUS: Main phases of the research completed successfully.
[Report]
TOPIC-6: CTA VESSEL SEGMENTATION
MOTIVATION: There exist many vessel segmentation algorithms for Computed Tomography Angiography (CTA) images. Currently, because of the complicated structure of the problem domain, the stable approach that works for variety of data sets without adjusting the algorithm parameters does not exist. Focusing on an approach that uses shape and intensity information in a balanced way should bring a novel solution to this problem.
GOAL: Development of a robust segmentation technique for CTA vessel data.
TECHNIQUES: Graph-cuts.
SOME RESULTS:
CTA image and the segmentation result, red represents the segmented vessel.
RESEARCH STATUS: Main phases of the research completed successfully.
[Report]