Evaluation of a semiautomatic contour detection approach in sequences of short-axis two-dimensional echocardiographic images.

Quantitative analysis of echocardiographic sequences has been limited by time-consuming and strenuous manual tracing approaches. To circumvent these disadvantages, we have developed the EchoCardiographic Measurement System (ECHO-CMS). ECHO-CMS employs the robust, model-based Minimum Cost Contour Tracking technique for semiautomatic detection of left ventricular (LV) endocardial contours in sequences of consecutive echocardiographic images. An evaluation study was carried out on 20 short-axis patient studies (10 transesophageal and 10 transthoracic studies), each consisting of 16 consecutive images covering approximately one cardiac cycle. The LV endocardial contours in the 320 images were analyzed both by manual tracing and semiautomatically. In addition, interobserver and intraobserver variabilities were determined for both techniques in two patients (32 images). Manual editing was required in only 57 (18%) of all 320 contours detected. Average processing time per patient for manual tracing was 25 minutes (of which 18 1/2 minutes was for drawing and corrections) and for semiautomatic tracing it was only 5 1/2 minutes (of which just 1 1/2 minutes was for corrections). Regression analysis showed excellent correspondence between manual and semiautomatic tracing: semiautomatic = 1.01 manual + 5.58%; r = 0.989; standard error of the estimate = 11.9% (n = 320 contours). Interobserver and intraobserver variabilities were smaller for semiautomatic than for manual tracing, although not in all cases statistically significant. In conclusion, semiautomatic LV short-axis endocardial contour detection by ECHO-CMS provides contours that are highly similar to those drawn by an expert; it is five to 10 times faster than manual tracing and reduces intraobserver and interobserver variabilities. This demonstrates that ECHO-CMS is a useful tool for clinical echocardiographic research studies.