[Accuracy and clinical applicability of the echocardiographic automated endocardial boundary detection system:...

Automated edge detection of endocardial borders is a new echo system that permits on-line delineation and tracking of blood-tissue interface and offers promise for measuring cyclic changes in cavity area and the assessment of left ventricular function on-line. Its accuracy has only been analyzed by linear regression and its applicability in unselected patients is not clearly established.

We analyzed the accuracy and clinical applicability of the system in 150 patients. Its accuracy was assessed not only by linear regression but also by bias analysis by comparing the results of the system with those obtained by manual tracing from parasternal short-axis and apical 4 chamber views.

We obtained satisfactory studies with this system in at least one of the two echo views in 66% of patients: 47% from 4-chambers view, 43% from parasternal short-axis view and 24.6% from both echo views. A visual semiquantitative assessment of left ventricular function could be performed in 88.6% of patients (p < 0.001). On-line and hand-traced left ventricular areas there well correlated, but fractional area change values from both methods correlated less closely. 95% limits of agreement between both methods were: 2.8 +/- 5.5 cm2 (end-diastolic area), 1 +/- 4.4 cm2 (end-systolic area) and 1.2 +/- 23% (fractional area change) in short-axis parasternal view; in apical 4-chambers view these limits were: 1.16 +/- 6.4 cm2,--1.1 +/- 6 cm2 and 7.3 +/- 16%, respectively.

Although the correlations between left ventricular areas from both methods were close the limits of agreement exceeded our acceptable range of reproducibility. Fractional area change showed only moderate correlations and a lack of agreement with off-line method. Poor image quality of the conventional echo still limits the clinical applicability of the current automated edge detection system.