The impact of real-time guiding by deep learning on test-retest variability of automated left ventricular systolic function measurements

Abstract

Abstract Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Norwegian University of Science and Technology, St. Olavs University Hospital, Central-Norway Health Authority. Background Left ventricular (LV) ejection fraction (EF) and left ventricular volumes are key parameters for characterisation of cardiac function. Test-retest variability of EF and LV end-diastolic volume (EDV) are dependent on image quality and analyses. Foreshortened LV recordings cause inaccuracies in estimation of EF and EDV. Real-time guiding of LV length using deep learning (DL) could improve standardization of image acquisition and reduce variability of automated analyses of EF (auto-EF) and EDV (auto-EDV). Purpose To study the impact of real-time feedback of LV length using a robust deep-learning (DL) software during echocardiography on variability of automated analyses of EF (auto-EF) and EDV (auto-EDV). Methods Patients with mixed cardiac pathology were consecutively included if they were in sinus rhythm without need for contrast. Three consecutive echocardiograms were performed; the first and second by two of three experienced sonographers (Sonographer 1 and 2) and the third (reference) by one of four cardiologists in random order, all affiliated to an EACVI accredited echocardiographic laboratory. All exams included the standard apical views, and the reference exams also included tri-plane recordings of the LV. All measurements were done retrospectively using automated algorithm for measuring. LV volumes were measured in four- and two-chamber views and averaged by the method of discs’ formula. LV EF was calculated as biplane measurements. The coefficients of variation (CV) between sonographers and cardiologists were compared for both LV EF and end-diastolic LV volume from biplane measurements. One-way ANOVA was used to estimate within-patient variation for automated and manual measurements. Results A total of 47 patients (47% women) were included with mean (SD) age 64 (15) years. There was no significant difference in CV for neither auto-EF nor auto-EDV using the DL tool (Table) compared to standard care. Manual and automated measurements of LV volumes and EF correlated well with R 0.91 and R 0.74, respectively (only data for EF shown in the figure). Conclusion Real-time guiding did not significantly reduce variation of automated analyses of LV volumes and EF when used by experienced sonographers. Further studies are needed to evaluate the effect of real-time guiding to optimize LV acquisition in inexperienced users.