Accurate left ventricle apex position and boundary estimation from noisy ventriculograms

Describes a methodology by which more accurate end-diastole and end-systole left ventricle boundaries can be automatically computed from the initial boundaries generated by a left ventricle pixel-based classifier. First, the end-diastole boundary is estimated from the boundary produced by a pixel-based classifier. Then the end-systole apex position is estimated from the end-diastole boundary. The end-systole boundary is then estimated using the end-systole boundaries produced by classifier and the estimated end-systole apex position. The methodology also permits a fusing of left ventricle boundary estimates produced by different techniques using a greedy algorithm. The authors used two different boundary estimation techniques having mean errors of 3.631 mm and 3.599 mm with standard deviation of 3.418 mm and 3.376 mm. The corresponding fused boundary has a mean error of 3.471 mm with standard deviation of 3.277 mm. The initial boundary produced by classifier had a mean error of 6.4 mm with an error of 8.5 mm in apex zone.