Probabilistic segmentation of myocardial tissue by deterministic relaxation

A recently developed probabilistic model for automatically segmenting regions of interest in abdominal CT (computer tomography) scans has been adapted to the task of segmenting myocardial tissue in cine-CT scans. A system has been implemented on relatively low-cost hardware which performs such segmentations. Special techniques have been developed to improve consistency and accuracy. Early results of testing this new modality are encouraging and promising. Extending the training set (even to the inclusion of aneurysms and other abnormal pathologies) actually improves segmentation performance in terms of accuracy and the number of iterations, required, contrary to initial expectations. In addition, using an extensible training set provides the means for folding in new results so that the system can learn from the addition of automated, as well as manual, segmentations. On the basis of observations from experimentation, new directions for future work have been identified