Leveraging deep learning algorithms for risk stratification and therapy planning in patients with aortic stenosis

Aortic valve stenosis (AS) is the most prevalent valvular heart disease in developed countries. Surgical aortic valve replacement (SAVR), the traditional standard of care for patients with severe symptomatic AS, is increasingly being complemented by transcatheter aortic valve implantation (TAVI). Several challenges (particularly regarding intervention timing and valve durability) need to be addressed before expansion to lower risk and younger patients including non-circular transcatheter valve deployment, crimp-induced leaflet damage, paravalvular leak, thrombosis, conduction abnormalities, prosthesis-patient mismatch, coronary occlusion, and aortic aneurysm. Several in vivo studies have reported the critical role of calcium morphology and distribution on AS severity and therapy outcome. Yet, there are no phenotype-related criteria for calcium as to which approach to use and when, and how, to balance benefits and risks. Abundant, inefficiently harvested information within the imaging data, if leveraged by the state-of-the-art routines of deep learning, could help characterize AS cases individually, predict clinical events, and plan for efficient therapy. In this study, a novel calcification scoring is proposed based on which a deep learning model is being developed to predict the clinical outcomes using the patient-specific CT images.