Search papers, labs, and topics across Lattice.
Artificial intelligence (AI) and deep learning (DL) are transforming cardiovascular diagnostics by enabling automated, scalable, and sensitive detection of heart disease from diverse data sources. This paper reviews contemporary AI and DL methods that improve diagnostic accuracy, reduce time-to-diagnosis, and enable population-level screening using electrocardiograms (ECG), echocardiography, coronary computed tomography angiography (CCTA), phonocardiograms (PCG), and wearable sensor signals. We present a taxonomy of model architectures including one-dimensional and two-dimensional convolutional neural networks, recurrent neural networks, transformers, and hybrid architectures, and we assess preprocessing, augmentation, and feature extraction techniques. Special emphasis is placed on clinically oriented evaluation: external validation across multi-centre cohorts, calibration, prospective clinical trials, and regulatory considerations for deployment. We synthesise evidence from recent large-scale studies showing that AI-enabled ECG algorithms and multimodal DL systems can detect structural heart disease, predict heart failure, and flag coronary artery disease with sensitivity and specificity that in some tasks approach or exceed expert clinicians. We discuss key barriers to clinical translation — data heterogeneity, label noise, algorithmic bias, interpretability, and workflow integration — and propose mitigation strategies including federated learning, uncertainty quantification, interpretable model design, and standardised reporting. Finally, we propose a focused research agenda prioritising prospective validation, transparent model sharing, clinician–AI co-validation frameworks, and implementation studies that measure impact on patient outcomes and health equity. Adoption of these technologies requires multi-stakeholder collaboration, robust post-deployment monitoring, and continuous model updating to ensure safety, generalisability, and sustained improvement in diagnostic outcomes for diverse patient populations worldwide. This review guides researchers and clinicians. Effectively.
