Аннотация
Cardiovascular diseases (CVDs) are the leading cause of mortality globally, with a disproportionate impact on low- and middle-income countries, including Kazakhstan and Central Asia [1], [2], [3]. Explainable Artificial Intelligence (XAI) has emerged as a promising approach to facilitate early CVD diagnosis by increasing model transparency, supporting clinical confidence, and enhancing decision support in preventive care settings [4], [5]. This systematic review evaluates 33 studies published between 2020 and 2025 that employed XAI approaches for early CVD diagnosis, following PRISMA 2020 guidelines. The studies were analyzed for dataset characteristics, modeling techniques, explainability methods, validation strategies, and clinical interpretability. Ensemblebased machine learning models, particularly Random Forest (36%), Decision Tree (30%), and XGBoost (27%), were most frequently utilized, achieving a median area under the curve (AUC) of 0.89 (range: 0.74-0.97). Deep learning methods, mainly Convolutional Neural Networks (21%) and Long ShortTerm Memory networks (12%), were primarily applied to ECG and imaging data. SHAP (54.5%) and LIME (27.2%) were the predominant explainability techniques, yet only three studies (9%) reported clinician-based validation of interpretability results. Despite methodological advances, 91% of the included studies relied on non-Central Asian datasets, highlighting a significant regional evidence gap. Limited access to local data, insufficient external validation, and minimal clinician involvement restrict the applicability of current XAI solutions to Kazakhstan and similar low-resource health systems. These findings underscore the need for standardized explanation evaluation frameworks and clinician-in-the-loop validation to support trustworthy and context-sensitive XAI deployment for early CVD diagnosis in Central Asia.