# Artificial intelligence and multimodal diagnostic approaches in cardiovascular disease

## TL;DR

This paper reviews how artificial intelligence can improve cardiovascular disease diagnosis by enhancing accuracy and efficiency compared to traditional methods.

## Contribution

The study systematically evaluates AI's clinical readiness and highlights barriers to adoption in cardiovascular diagnostics.

## Key findings

- AI models achieved over 90% accuracy in cardiac imaging tasks like ventricular dysfunction detection.
- Deep learning models reached 0.99 AUC for predicting atrial fibrillation and ischemic heart disease.
- Biomarker-based ensemble models achieved over 95% diagnostic accuracy when combining proteomic and clinical data.

## Abstract

Evaluate the impact and clinical applicability of artificial intelligence (AI) models in cardiovascular diagnosis, assessing their potential to improve diagnostic accuracy, operational efficiency, and reliability compared with conventional methods.

Methods. A critical review of the recent literature was conducted, encompassing retrospective studies, multicenter trials, and external validations that employed machine learning and deep learning algorithms applied to imaging modalities, electrocardiographic and phonocardiographic signals, as well as clinical and proteomic biomarkers.

Evidence indicates that in cardiac imaging, automated segmentation and ventricular dysfunction detection achieved accuracy metrics exceeding 90%, suggesting readiness for clinical integration. In cardiac signals, deep learning models demonstrated area under the ROC curve values of approximately 0.99 for predicting atrial fibrillation and ischemic heart disease, further supported by explainability techniques. Regarding biomarkers, ensemble models achieved diagnostic accuracies above 95%, and the integration of proteomic and clinical data substantially enhanced predictive performance. Nonetheless, decreased performance in external validations, limited generalizability to heterogeneous populations, and clinicians’ reluctance due to insufficient explainability remain major barriers.

Artificial intelligence in cardiovascular diagnostics holds transformative potential by improving accuracy, reducing interobserver variability, and expanding access in resource-limited settings. However, its consolidation into routine practice requires robust multicenter validations, seamless interoperability with clinical workflows, and strengthened explainability, prerequisites for incorporation into clinical guidelines and precision medicine strategies.