TRUSTWORTHY MULTIMODAL AI FOR CLINICAL DECISION SUPPORT: SAFETY, EXPLAINABILITY, AND RELIABILITY

Abstract

Introduction: Multimodal artificial intelligence (AI) is a subset of artificial intelligence that is increasingly utilized in support systems for clinical decisions by incorporating various healthcare data. Nevertheless, its validity and clinical use is hampered by safety issues, explainability and reliability concerns.

Aim: This study aims to examine the trustworthiness of multimodal AI in clinical decision support by analyzing safety, explainability, and reliability as core dimensions.

Methodology: A qualitative narrative review was conducted using secondary data from recent peer-reviewed studies and policy literature. Thematic analysis was used to generalize the findings based on three main areas, which are safety, explainability, and reliability. A systematic review and analysis of 20 relevant studies were performed.

Findings: The findings show that multimodal AI enhances decision-making because it combines heterogeneous information but presents complex risks including imbalance in datasets, absence of modalities, and lack of external validation. Explainability methods promote clinician insights, but alone are not adequate to warrant trust. Reliability is a significant issue as it is affected by variation in clinical settings, the change of time, and discrepancies in the data. Combined analysis reveals that trustworthiness is the interaction of all three dimensions backed by governance and human control.

Conclusion: Reliable multimodal AI demands a holistic solution that entails technical, explanatory, and realistic validation. Future studies can concentrate on standard evaluation models and clinical implementation models