Multimodal AI in Cardiolog y: a critical review of diagnostic performance and clinical applicability

Ramos-Zaga F. · Revista de la Federacion Argentina de Cardiologia (2026) · DOI: 10.63600/y76zfg77

Resumen

Introduction: cardiovascular diseases continue to impose a substantial clinical burden and call for diagnostic strategies capable of integrating imaging, biomedical signals, and clinical data. In this context, artificial intelligence has shown promise in supporting cardiac segmentation, tissue characterization, and the detection of arrhythmia, ischemia, and valvular disease; however, its performance must be interpreted considering validation conditions. Objective: to critically assess the reported diagnostic performance of artificial intelligence systems applied to cardiovascular imaging and biomedical signal analysis, with particular attention to validation settings and methodological constraints that may hinder clinical translation. Methods: a narrative review with a systematic orientation was conducted in accordance with PRISMA 2020 guidelines. Searches were performed in Scopus, Web of Science, and PubMed, including empirical studies published between 2021 and 2024. Of 872 records identified, 17 studies met the inclusion criteria. Results: imaging-based studies reported high performance in echocardiographic segmentation, with Dice coefficients exceeding 0.90 in selected tasks, although limitations emerged in the presence of low image quality, small sample sizes, or limited external validation. Signal-based studies achieved AUROC values above 0.90 in several applications, yet showed variable specificity and notable performance drops in external cohorts. Conclusion: the available evidence points to meaningful diagnostic potential, shaped by sample heteroge-neity, overfitting, selection bias, reliance on limited metrics, and regulatory shortcomings. Clinical translation is likely to depend on robust prospective validation, independent external testing, and ongoing monitoring.

Instituciones: Universidad Privada del Norte