Incremental Learning For Artificial Intelligence İn The Prediction And Classification Of Atherosclerotic Cardiovascular Disease

Abstract

Objective: Atherosclerotic Cardiovascular Disease (ASCVD) remains a leading cause of morbidity and mortality worldwide and represents a substantial public health burden. Artificial Intelligence (AI) and Machine Learning (ML) are increasingly integrated into ASCVD diagnosis, prediction, and management. This review aims to evaluate the role and potential of Incremental Learning (IL) in improving ASCVD risk prediction and classification, addressing the limitations of static ML models.

Methods: This narrative review was developed in accordance with the SANRA criteria. Current AI/ML models commonly applied in ASCVD—such as Logistic Regression, Random Forests, and Convolutional Neural Networks—were examined with a focus on their reliance on static datasets and vulnerability to performance degradation caused by concept drift. The challenges of data heterogeneity, quality issues in Electronic Health Records (EHRs), and the computational burden of repeated full-model retraining were analyzed. Literature on Incremental Learning algorithms was reviewed to assess their applicability in continuously evolving clinical environments.

Results: Incremental Learning enables AI models to integrate new data over time while preserving previously acquired knowledge, thus enhancing adaptability in the context of evolving ASCVD risk factors, diagnostic criteria, and clinical guidelines. The approach supports more accurate and individualized risk prediction and reduces computational overhead. However, catastrophic forgetting—loss of earlier knowledge during adaptation to new data—emerged as a key limitation. Advanced IL strategies, including experience replay, have shown potential in mitigating this issue, but further validation is required for clinical implementation.

Conclusion: Incremental Learning offers a forward-looking, adaptive framework for ASCVD prediction and classification, outperforming traditional static ML models in dynamic and heterogeneous data settings. Despite its advantages, ensuring model stability and preventing catastrophic forgetting remain essential for safe and reliable clinical deployment. Future research should prioritize methodological refinements and robust validation to support the integration of continuously learning AI systems into ASCVD management.