Artificial Intelligence for Early Detection of Subclinical Atherosclerosis: A New Frontier in Preventive Cardiology

Abstract

Cardiovascular disease (CVD), the leading cause of mortality worldwide, often progresses silently for years until manifesting as symptomatic ischemia or infarction (1, 2). Subclinical atherosclerosis—defined here as a coronary artery calcium (CAC) score > 0, carotid intima-media thickness (CIMT) above the 75th percentile, or the presence of non-obstructive plaque—typically remains undetected until significant damage occurs. While conventional diagnostic tools like coronary CT angiography (CCTA) are resource-intensive and operator-dependent, artificial intelligence (AI) offers a transformative opportunity to bridge this gap using routine clinical and imaging data. 

Recent advances in deep learning (end-to-end models) and radiomics (hand-crafted feature extraction) have demonstrated remarkable precision in quantifying plaque burden. For instance, AI-enhanced CCTA algorithms now enable fully automated CAC scoring with near-human accuracy (r = 0.97). Validated against expert readers in cohorts using multiple cardiac and chest CT protocols, these tools substantially reduce analysis time (1, 2). Similarly, convolutional neural networks (CNNs) applied to carotid ultrasound have achieved >90% sensitivity in identifying plaque morphology and stroke risk (3). Machine-learning models integrating non-imaging predictors like lipid profiles and ECG parameters further refine risk estimation, outperforming traditional ASCVD scores (4, 5). 

Future Directions and Regional Implementation:

In South Asia, where the background CVD burden is exceptionally high, the transition to proactive prevention is critical. While CCTA access is limited in rural settings, simple and scalable AI approaches—such as ECG-based models or smartphone-based retinal imaging—offer a practical pathway for community-level screening. Furthermore, opportunistic algorithms can extract CAC scores from routine chest CT scans obtained for non-cardiac reasons, expanding surveillance without additional radiation or cost (1, 2). 

However, the implementation of AI in cardiovascular screening must address several critical barriers: 

• Overdiagnosis and False Positives: Incidental low-risk findings may generate patient anxiety and lead to unnecessary, invasive downstream testing.
• Infrastructure and Cost: Implementation is often hindered by limited digital infrastructure, the high cost of AI tools, and the need for specialized workforce training in low-resource settings.
• Regulatory and Medico-Legal Aspects: Clear frameworks are required to establish responsibility when AI output conflicts with clinician judgment.
• Algorithmic Bias: Models trained primarily on Western datasets may underperform in diverse populations due to differing plaque morphology; therefore, future research must prioritize multi-ethnic dataset inclusion.

We propose a three-year roadmap to conduct a prospective multicenter study in South Asia. This study will evaluate AI-assisted detection against standard assessments, focusing on primary clinical outcomes such as statin prescription rates and long-term event prevention. By leveraging AI, cardiology can shift from reactive intervention to proactive prevention, enhancing global health through precision and accessibility