Left ventricular viability maps: Fusion of multimodal images of coronary morphology and functional information

Our objective is to create a patient-specific map explicitly combining coronary territories and myocardial viability. This continuous model would adapt to the patient and allow the study of groups of coronary unavailable with standard models. After having identified loss of viability, the clinician would use this model to infer the most likely obstructed coronary artery responsible for myocardial damage. Visualization of the loss of viability along with coronary structure would replace the physician's task of mentally integrating information from various sources. Anatomic (or structural) information of the heart is available with Computer Tomography Angiography (CTA). A contrast agent is administered through the arm and spreads until it reaches the coronaries. A CT scan is performed at that moment and coronaries appear bright on the resulting images. The surfaces of the myocardium (epicardium and endocardium) are identifiable and coronary anatomy of individual patients can be determined by experienced physicians. Once the coronary arteries and LV surface is recovered, coronary territories can be computed over the epicardial surface using the Dijkstra algorithm. This method finds the shortest path from a source point, a point on a coronary projected onto the LV surface, to all other point of the surface. Each point on the surface is then assigned a label corresponding to the closest coronary.;Fusion of both modalities (CTA and DEMR) is needed in order to guarantee the correspondence of structure and function in the final map. We chose to accomplish an initialization of registration by Principal Component Analysis (PCA) which determines and aligns the directions of maximal variance of two clouds of points (the surface points of the LV in both modalities). As a result, the LV in both DEMR and CTA images are roughly lined up in 3D.;This rough estimation of the registration suffers from two drawbacks. First, it is a rigid registration and the heart deforms non-rigidly. Second, the LV is rotationally symmetric in the short axis plane so this aspect of the alignment may not be correct. Once the DEMR and CTA images have been roughly aligned by PCA, we can proceed to a more precise registration and exploit the intensity information contained in each available volume with mutual information. Thin plate spline (TPS) fitting consists of finding a deformation according to given source and target landmarks. We apply it to obtain an elastic registration of myocardial surfaces. Once a 3D registration has been completed, a projection can be performed from 3D to 2D. Thus, a complete model is obtained where coronary territories are overlaid with viability information for a specific patient.;Clinical evaluation was done over 6 patients. Coronary territories could be computed for four of them and viability information was available for four others. Both structural and functional information was available for two patients. Validation for our computed coronary territories was done by a comparison with the Green Lane method. In this method, given a coronary angiogram of the patient, the coronary arteries and branch vessels are transcribed onto a standard diagram by an expert cardiovascular imager and resulting coronary territories are drawn by an expert. By employing the Green Lane method to create ground truth we were able to quantify the correctness of both our territory map as well as the 17-segment model. Overall, the average percent correlation between our CTA-derived maps and the ground truth was 87%. The 17-segment model comparison resulted in a correlation of 79%. Registration was evaluated qualitatively.;Cardiac function may be inferred from Delayed Enhancement Magnetic Resonance (DEMR) images. In this protocol, a contrast agent is administered to the patient and image acquisition is done after a known delay corresponding to the wash out rate of healthy myocardial cells. After that delay, any cells still retaining the contrast agent are considered dead. This phenomenon is hypothesized to be caused by the slow wash out rate of dead cells. Thus, bright areas of the myocardium indicate a possible site of scar.;We achieved our goal of creating a patient-specific coronary territory map fused with a viability map which presents the functional as well as structural information of interest to the clinician for the diagnosis of CHD. While the presented results are an improvement over the 17-segment model, some criticism can be formulated for each step of our method and there is place for improvement at different levels. (Abstract shortened by UMI.)...