Reliability of three computer methods in the analysis of ECG-gated radionuclide left ventriculography: interrecording, interobserver and intraobserver variability.

Three different computer methods for analysis of systolic and diastolic left ventricular function (ejection fraction, peak ejection rate, time to peak ejection rate, peak filling rate, time to peak filling rate, duration of fast filling phase, and fast filling fraction) as derived from ECG-gated radionuclide cardiography were compared in 30 patients with various diseases. The patients had two gamma camera recordings of the left ventricle performed immediately following one another during radionuclide (99mTc) equilibrium (3 x 10(6) counts, 16 frames/cycle, 64 x 64 pixels). Mean ECG R-R interval of the patients remained unchanged from first to second recording. The three computer methods were: (1) end-diastolic (ED) region of interest (ROI) analysis based on manually defined ED-ROI; (2) multi (M) ROI, manually defined ROI for each frame; and (3) semiautomatic (SA) ROI, ROI for each frame defined by an SA edge detection technique. With the 16 frame points as nodes, a 160-point time-activity curve was constructed for each of the three methods by use of a spline function. A tailored multiway analysis of variance showed that the M-ROI method had the highest interindividual range of values of the function parameters and the smallest interrecording, interobserver, and intraobserver variabilities. In theory this implies a better diagnostic sensitivity and specificity for the M-ROI method as compared with the other two methods.