Impact of real-time feedback by deep learning during echocardiography on test-retest variability of left ventricular systolic function measurements.

Background: Left ventricular (LV) foreshortening is a common problem within echocardiography and cause inaccuracies in estimation of LV ejection fraction (EF) and end-diastolic volume (EDV). Purpose: To evaluate the impact of real-time feedback of LV length, using a robust deep learning (DL) tool during echocardiography, to reduce test-retest variability in assessment of LV EF and EDV. Method: Included patients scheduled for echocardiography at the “Echo Lab” at St. Olavs hospital. Each patient had 3 exams. Study population: 88 patients included, 45% were women and the mean age was 63 years of age. Results: foreshortening was significantly reduced by a mean of 2.4 mm after the introduction of DL tool. Alignment of the mitral annulus was numerically improved. But, there was no significant difference in coefficient of variation for neither LV EF nor EDV using the DL tool. Conclusion: To our knowledge, this is the first time a DL tool was used for real-time feedback of LV length, and when used by experienced sonographers it significantly reduced foreshortening and provided more standardized recordings. However, it did not result in significant improvement in test-retest variation of LV EF and LV EDV.