Additional prognostic value of fully-automatic Global Longitudinal Strain using machine learning

Abstract Funding Acknowledgements Type of funding sources: None. Background Several studies described the independent prognostic value left ventricular global longitudinal strain (GLS) using cardiovascular magnetic resonance (CMR) to predict events. However, potential interest GLS a fully automatic method assessed at rest during stress CMR exam was not well established. Aim To assess all-cause death machine learning algorithm without human correction in consecutive patients referred for CMR. Methods Between 2016 and 2018, all were included followed occurrence death. A trained validated unseen (MAGNETOM Aera Skyra, Siemens Healthcare, Erlangen, Germany) from long-axis cine images acquired rest. The combines multiple deep networks detection segmentation with an active contours approach. Cox regressions performed determine GLS. Results Among 9,883 who underwent between 2018 retrospectively included, successfully computed 9,638 (97.5%) (67% male, mean age 66±12 years). total 510 (5.3%) deaths observed median (IQR) follow-up period 4.5 (3.7–5.2) years. GLS, presence inducible ischemia late gadolinium enhancement (LGE) significantly associated (hazard ratio, HR: 1.22 [95% CI, 1.17–1.26]; 2.23 1.61–3.10]; 2.04 1.41–2.95], respectively, p<0.001). After adjustment traditional risk factors, LGE, predictor higher incidence (adjusted 1.14 1.08–1.20]). Automatic showed incremental compared LGE (C-statistic improvement: 0.05; NRI=0.146; IDI=0.244; Conclusions measured has above other parameters.