Segmentation of Echocardiographic Image Sequences Using Spatio-temporal Information

This paper describes a new method for improving border detection in image sequences by including both spatial and temporal information. The method is based on three dimensional quadrature filters for estimating local orientation. A simplification that gives a significant reduction in computational demand is also presented. The border detection framework is combined with a segmentation algorithm based on active contours or ’snakes’, implemented using a new optimization relaxation that can be solved to optimality using dynamical programming. The aim of the study was to compare segmentation performance using gradient based border detection and the proposed border detection algorithm using spatio-temporal information. Evaluation is performed both on a phantom and in-vivo data from five echocardiographic short axis image sequences. It could be concluded that when temporal information was included weak and incomplete boundaries could be found where gradient based border detection failed. Otherwise there was no significant difference in performance between the new proposed method and gradient based border detection.