Trustworthy AI in Digital Health: A Comprehensive Review of Robustness and Explainability

AI systems in healthcare require multiple safeguards beyond accuracy alone. The paper emphasizes that trustworthy AI depends on five key dimensions—robustness, explainability, fairness, accountability, and privacy—that must be addressed throughout the entire lifecycle from initial problem design through deployment and human interaction.

Explainability is critical for clinical adoption. The review highlights that doctors and patients need to understand why an AI system makes specific recommendations, particularly in high-stakes decisions affecting patient care. Without explainable AI methods, healthcare providers cannot verify whether the system's reasoning is sound or detect potential errors.

Real-world healthcare data creates unique challenges. The paper identifies that AI models often fail when deployed in actual clinical settings due to "distributional shifts"—differences between training data and real patient populations. The review addresses techniques for improving robustness when healthcare data is scarce and variable across different hospitals or patient groups.

Trust requires accountability across the entire AI pipeline. Rather than focusing only on the final model, the paper stresses that trustworthiness depends on responsible choices made during data collection, model selection, and deployment decisions—meaning organizations must take responsibility for how AI systems are developed and used, not just their performance metrics.

Different healthcare applications have distinct trust needs. The review demonstrates that intensive care, neonatal health, and metabolic health applications each require tailored approaches to robustness and explainability, suggesting that one-size-fits-all AI solutions are insufficient for responsible healthcare deployment.