An Automatic Left Ventricle Segmentation on Echocardiogram Exams via Morphological Geodesic Active Contour with Adaptive External Energy

This work proposes a new adaptive approach to left ventricle segmentation based on a non-parametric adaptive active contour method called Fast Morphological Geodesic Active Contour (FGAC) combined with adaptive external energy via deep learning model. The evaluation methodology considered echocardiogram exams obtained from volunteers. Beyond the manual segmentations made by two specialists medical as ground truth. The new approach is compared with three other segmentation methods, also based on the active contour method: pSnakes, radial snakes with derivative (RSD), and radial snakes with Hilbert energy (RSH). The FGAC combined with adaptive external energy showed better Precision (99.53%, 99.72%) against RSD (99.46%, 99.68%), RSH (99.51%, 99.71%) and pSnakes (99.52%, 99.72%). Besides, it achieved a relevant Jaccard similarity index (67.40%, 62.02%), and promising accuracy (98.64%, 98.46%). Even though the metrics differences are low, the proposed approach is fully automatic. Therefore, these results suggest the potential of the proposed approach to aid medical diagnosis systems in echocardiology.