A Review of Trust-Aware Explainable Federated Multimodal Learning for Early Cardiovascular Disease Prediction

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Cardiovascular disease (CVD) is the largest cause of death worldwide, and artificial intelligence can now detect it earlier than traditional methods. The difficulty is that the data needed to train a reliable model is held separately by individual hospitals and is protected by privacy law. Federated learning (FL) answers part of this problem by training a shared model without moving patient data out of each hospital. This review examines thirteen studies published between 2022 and 2026 and organises them around the four capabilities a usable CVD prediction system needs: multimodal data preparation, trust-aware aggregation, federated multimodal fusion, and explainable privacy-preserving decision support. The evidence is presented in table form for ease of comparison. Each capability has now been demonstrated on its own. Cross-attention fusion of ECG and clinical data outperforms ECG alone; reputation-weighted aggregation keeps a federated model accurate when some clients are unreliable; adaptive differential privacy reduces membership inference attack success to near chance; and SHAP and LIME explanations can now be scored for stability rather than simply displayed. No published framework, however, combines all four and evaluates them on cardiovascular data under one protocol. This review sets out that gap and describes the four-phase framework intended to close it.