Deep Learning in Cardiovascular CT Imaging: Evolution, Trends, and Clinical Translation from 2020 to 2025

Abstract

Cardiovascular diseases remain the leading cause of mortality worldwide, necessitating advanced diagnostic approaches. Deep learning has revolutionized cardiac computed tomography (CT) imaging analysis over the past five years, transforming from experimental algorithms to clinically validated tools. This review examines the architectural evolution from convolutional neural networks to transformer-based models, analyzing their application across anatomical segmentation, coronary artery analysis, and functional risk assessment. We synthesize findings from 15 high-impact studies published between 2020 and 2025, documenting performance improvements in cardiac chamber segmentation (Dice coefficients 0.88-0.95), stenosis detection (sensitivity 90-97%), and mortality risk prediction (C-index 0.70-0.82). Technical challenges including limited annotated datasets, cross-scanner generalization, and regulatory barriers continue to impede widespread clinical adoption. Emerging foundation models and multimodal integration represent promising directions for next-generation cardiovascular AI systems.