Deep learning methods such as convolutional neural networks dominate recent AI-driven ECG research, though dataset limitations and lack of external validation remain key barriers to clinical translation.
Systematic Review (n=83)
This systematic review highlights that while deep learning models like CNNs dominate AI-driven ECG analysis, real-world clinical adoption is currently limited by a lack of external validation and poor model interpretability.
Cardiovascular diseases (CVDs) remain one of the primary global health challenges, emphasizing the necessity for fast and accurate diagnostic methods. Among the available diagnostic tools, electrocardiography (ECG) is commonly used because of its non-invasive nature, accessibility, and low cost for detecting cardiac abnormalities. The integration of Artificial Intelligence (AI) techniques has significantly improved ECG-based diagnosis by automating ECG interpretation and supporting clinical decision-making. Although several surveys have reviewed AI in CVD diagnosis, most take a broad perspective across imaging, health records, and wearable data. A focused and systematic synthesis of AI-driven electrocardiogram (ECG) analysis is still lacking. This review addresses this gap by systematically analyzing 83 studies published between 2016 and 2025. A structured methodology was applied to identify and evaluate studies in terms of datasets, AI models, diagnostic tasks, and evaluation practices. The findings show that deep learning methods such as convolutional neural networks dominate recent research, with PTB-XL and MIT-BIH as the most widely used datasets. However, dataset limitations, lack of external validation, and limited attention to interpretability remain key barriers to clinical translation. By providing a focused, systematic, and comparative synthesis, this review extends existing surveys and highlights research gaps, offering a roadmap to guide future AI-driven ECG-based cardiac diagnostics toward real-world adoption.
Houssein et al. (Sun,) conducted a systematic review in Cardiovascular diseases (n=83). Artificial Intelligence (AI) techniques for ECG-based diagnosis was evaluated. Deep learning methods such as convolutional neural networks dominate recent AI-driven ECG research, though dataset limitations and lack of external validation remain key barriers to clinical translation.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: