To synthesize recent advances in AI applications for improving donor-recipient matching, non-invasive rejection surveillance, and pathological diagnosis in heart transplantation, focusing on studies published between October 1, 2020, and October 1, 2025.
Key Findings:
3D-Convolutional Neural Networks (3D-CNNs) enabled accurate total cardiac volume measurement for donor-recipient matching, highlighting the need for methodological rigor.
Machine learning models identified non-linear risk factors for postoperative adverse events, outperforming traditional models.
AI models integrating non-invasive biomarkers showed high diagnostic accuracy for rejection, potentially reducing unnecessary EMB procedures.
AI improved sensitivity for detecting high-grade acute cellular rejection from 39.5% to 74.4% compared to manual assessments.
Interpretation:
AI has the potential to significantly enhance heart transplantation practices by optimizing matching, enabling non-invasive monitoring, and improving diagnostic accuracy, but current evidence is limited by methodological issues that need to be addressed.
Limitations:
Most studies were single-center and retrospective, limiting generalizability.
Small sample sizes and lack of independent external validation were common.
Diverse populations were often not represented, affecting the applicability of findings.
Conclusion:
While AI shows promise in heart transplantation, further research is needed to validate findings through prospective, multi-center studies and to standardize reporting practices for clinical integration, addressing current methodological limitations.
