AI-driven hybrid framework for the generation of population-specific cardiovascular virtual cohorts

Abstract

In silico clinical trials (ISCTs) offer a promising approach to accelerate the commercialization of cardiovascular devices. In this study, we present a novel synthetic data generation framework for generating clinically relevant cardiovascular geometries for testing the devices on a large population. A hybrid framework comprising a combination of artificial intelligence (AI) and statistical shape modelling (SSM) approach has been proposed for the generation of distinct cardiovascular geometries (aortic valves and coronary arteries) along with the pathologies. Synthetic aortic valves are generated for male and female populations, respectively. The evaluation of the anatomical features of these synthetic valves exhibits clear evidence of clinical realism. In addition, valves with unique anatomical variations are generated, and assessment of sub-population-specific features confirms the effectiveness of the framework. The framework further extends to the generation of another sub-population for synthetic coronary arteries. The study presents a widely applicable and scalable hybrid synthetic data generation framework for the generation of distinct cardiovascular geometries for specific populations. The results evidently show that the framework is highly suitable for the generation of anatomically precise and population-specific virtual cohorts to support future ISCTs.