Automated detection of left ventricular borders from intracardiac ultrasound images

An automated method for segmentation of epicardial and endocardial borders from intracardiac ultrasound (ICUS) images is reported. Optimal borders are detected within a single interactively defined elliptical region of interest using graph searching. A priori knowledge about the ultrasound echo responses arising from the myocardial borders as well as information about heart wall morphology is incorporated in the authors´ graph search. The detected epicardial borders serve as the shape model during the detection of the endocardial borders. The method was applied to 26 ICUS images acquired in vivo in 2 dogs. The heart wall borders were successfully detected in all images. Good correlation was obtained between computer-detected and observer-defined left ventricular (LV) cavity areas and LV areas (r=0.99, y=0.98x+0.02; r=0.99, y=0.97x+0.86; respectively). Epicardial and endocardial borders were correctly positioned with a root-mean-square border positioning error of 0.96/spl plusmn/0.28 mm. The authors´ results demonstrate the feasibility of automated border detection in ICUS images.