JMIR Publications

ABSTRACT

Background:

Human Digital Twins (HDTs) represent a revolutionary leap in the field of cardiac medicine. They have been made possible by the fusion of personal physiological information with the application of artificial intelligence concepts in modeling cardiac activity. However, studies have shown that the field of AI-driven cardiovascular HDT studies has become fragmented, lacking a holistic framework for research methodology.

Objective:

This review aims to consolidate global evidence on AI-driven cardiovascular HDTs by systematically classifying the 73 included full-text studies into seven application domains, mapping regional contributions and highlighting regional leading roles alongside notable equity gaps, identifying the most frequently employed modelling architectures, and compiling benchmark datasets with evaluation metrics to guide reproducible modelling.

Methods:

Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR) guidelines for systematic and scoping reviews, literature published between 2018 and November 2025 was retrieved from a database (IEEE Xplore, ScienceDirect, SpringerLink, ACM DL, and Web of Science). The search was conducted using word combinations such as “AI models”, Human Digital Twins”, and “Cardiovascular Medicine”. Only those studies that reported Machine learning and Deep learning for Cardiovascular HDTs were included. As such, the review included 73 full-text studies that met the inclusion criteria after duplicate removal and multi-stage screening, and these were analysed through qualitative and quantitative synthesis.

Results:

This review synthesizes the studies that reported the use of AI models for various applications of HDTs in cardiovascular disease care. The findings show a marked rise in publications after 2021, with a strong concentration of studies in 2024 and 2025, reflecting the field’s rapid acceleration. Most studies were centered on patient-specific modeling, representing 40.5%, while risk prediction and physiological monitoring were also major areas of application. Regionally, Asia dominated the research landscape with 29 studies (44.59%), driven primarily by India with 13 studies (44.83%) and China with 11 studies (37.93%). Regarding modeling approaches, Convolutional Neural Networks (CNNs) accounted for 27.54%, while Recurrent Neural Networks and LSTM models accounted for 8.70%. Multimodal data integration was evident in 26 studies (35.14%), with imaging modalities used in 30 studies (40.54%). Across the included studies, performance evaluation relied heavily on accuracy-based metrics, with 23 studies (31.08%) reporting Accuracy, ROC-AUC, or F1-Score as key indicators. However, external validation was limited to fewer than 10% of the studies, underscoring a continued translational gap between technical validation and real-world clinical assessment.

Conclusions:

This review clearly identifies the great promise of artificial intelligence and machine learning (AI/ML) algorithms for addressing challenges in patient-specific modeling, risk evaluation, and cardiovascular diagnostics. The review also discusses various concerns associated with the standardization of benchmarks and evaluation measures, which are critical to the use of cardiovascular HDTs.