Comparison of fully automated and manual ejection fraction calculations: validation and pitfalls

Resting multigated blood-pool studies were performed on 61 patients without arrhythmias and data were simultaneously acquired to two computer systems. Using one computer, manual ejection fraction (EF) was calculated by two trained observers. EF was also calculated from the other computer using a commercially available fully automated program; quality control (QC) images were routinely obtained to evaluate correct left-ventricular center location, background region assignment, or gross edge mispositioning. When errors were noted, the automated analysis was reprocessed with operator intervention. Forty-eight of the 61 studies (78%) produced adequate QC images. Operator redefinition of the left ventricular center and background of the 13 QC failures raised the automated success level to 92%. Correlation of the manual EF by two observers was excellent (r = 0.969). The automated EF measurements correlated well with the average observer's (r = 0.898). An improved version of software reduced the QC failures from 13 to 10. Operator assisted automated processing gave the success rate of 94%. The remaining 6% of patients required manual processing to obtain a correct ejection fraction. Our normal range for manual EF is greater than or equal to 0.50 and our corresponding normal range for this group of patients using the automated program was greater than or equal to 0.44. Based on greater than or equal to 0.44, four patients with a low manual EF had a normal automated EF. All four patients had cardiac disease but there was no evidence of abnormal cardiac function. The fully automated program provided good correlation with manual EF and can remove some of the subjectivity in manual edge determination. However, QC images must be carefully examined and the normal EF range for the automated program must be determined.