Two automatic calibration algorithms for left ventricle boundary estimation in X-ray images

Non-homogeneous mixing of the dye with the blood in the left ventricle and low contrast in the apex zone causes pixel-based classifiers to yield boundaries which are not close physician traced boundaries. They have a systematic positional and orientational bias, with under-estimation in the apex zone. This paper develops two calibration algorithms, the identical coefficient and the independent coefficient. These algorithms transform the two sets of given training boundaries: physician traced and the classifier, to yield the off-line parameters, depending upon the number of partitions of the database. Vertices of the left ventricle boundary are then estimated on-line by applying these transformed parameters. The performance of the calibration system based on the polyline distance metric yields a mean error of 3.7 and 3.6 millimeters for above algorithms over 6×10 ⁴ vertices in the data base of 291 patient studies. Both calibration algorithms remove the bias and reduce the boundary error in the apex zone. For end-diastole frame the system reduces the error by 8.5 millimeters in the apex zone over the pixel-based classifier boundaries produced by image processing algorithms