Automatic detection of left ventricular borders on electron beam CT sequential cardiac images using an adaptive algorithm.

Detection of myocardial borders on sequences of electron beam CT images is carried out using an adaptive segmentation algorithm developed to enhance dynamic analysis of cardiac function. Adaptivity is based on description of the myocardial borders from the mean and standard deviation of the grey level and gradient distributions on each image of the sequence. Comparison of segmentations from five experimentators with automatically determined borders on a set of 416 endocardial and epicardial contours indicated differences between automatic and manual tracing very close to differences due to inter-observer reproducibility.