Clinical utility of automated assessment of left ventricular ejection fraction using artificial intelligence-assisted border detection

Background: Ejection fraction (EF) calculated from 2-dimensional echocardiography provides important prognostic and therapeutic information in patients with heart disease. However, quantification of EF requires planimetry and is time-consuming. As a result, visual assessment is frequently used but is subjective and requires extensive experience. New computer software to assess EF automatically is now available and could be used routinely in busy digital laboratories (>15 000 studies per year) and in core laboratories running large clinical trials. We tested Siemens AutoEF software (Siemens Medical Solutions, Erlangen, Germany) to determine whether it correlated with visual estimates of EF, manual planimetry, and cardiac magnetic resonance (CMR). Methods: Siemens AutoEF is based on learned patterns and artificial intelligence. An expert and a novice reader assessed EF visually by reviewing transthoracic echocardiograms from consecutive patients. An experienced sonographer quantified EF in all studies using Simpson's method of disks. AutoEF results were compared to CMR. Results: Ninety-two echocardiograms were analyzed. Visual assessment by the expert (R = 0.86) and the novice reader (R = 0.80) correlated more closely with manual planimetry using Simpson's method than did AutoEF (R = 0.64). The correlation between AutoEF and CMR was 0.63, 0.28, and 0.51 for EF, end-diastolic and end-systolic volumes, respectively. Conclusion: The discrepancies in EF estimates between AutoEF and manual tracing using Simpson's method and between AutoEF and CMR preclude routine clinical use of AutoEF until it has been validated in a number of large, busy echocardiographic laboratories. Visual assessment of EF, with its strong correlation with quantitative EF, underscores its continued clinical utility.