Automated real-time analysis of intraoperative transesophageal echocardiograms

Although transesophageal echocardiography (TEE) produces real-time images depicting left ventricular (LV) filling and ejection, the quantitative analysis of these images has been too time consuming to be of practical value in the operating room. Therefore, the authors investigated whether a new automated border detection system (ABD) could track the endocardial border continuously and compute the cross-sectional area of the LV cavity.

Using data from 25 patients who were monitored with TEE as part of their routine clinical care, the authors compared ABD estimates of LV end-diastolic area (EDA in square centimeters), end-systolic area (ESA in square centimeters), and fractional area change (FAC) with the laboratory measurements made independently by an expert.

ABD slightly underestimated EDA (10.7 +/- 1.0 vs. 11.2 +/- 1.0 cm2) and slightly overestimated ESA (5.6 +/- 0.7 vs. 4.8 +/- 0.6 cm2, mean +/- standard error). However, when ABD tracking of the endocardial border was judged as "good" or "excellent" (84% of the patients at end diastole and 72% at end systole), the limits of agreement between ABD and the expert's findings were within the limits expected for two experts. By contrast, ABD significantly underestimated FAC (0.44 +/- 0.03 vs. 0.56 +/- 0.03) and the limits of agreement between ABD and the expert were more than twice as great as expected for experts, even when ABD performance was judged as "excellent."

The authors conclude that, when ABD appears to be performing adequately, it underestimates LV FAC, but provides valid real-time estimates of LV EDA and ESA. Thus, it warrants further evaluation as a potentially powerful clinical and research tool.