Explainable AI for clinical and remote health applications: a survey on tabular and time series data

XAI is underutilized in healthcare data types that matter most: While explainable AI has received significant attention in computer vision and natural language processing, the survey finds that XAI methods are "not commonly analysed" on tabular and time series data—the primary data types used in clinical and remote health applications, representing a critical gap in responsible AI deployment.

Complex AI systems dominate healthcare despite transparency needs: The paper identifies a fundamental tension in healthcare AI: "the best performing AI systems are often too complex to be self-explaining," yet explainability becomes "even more critical when dealing with sensitive and personal health data," highlighting the need to prioritize interpretability alongside accuracy in medical contexts.

Multiple validation approaches are essential for trustworthy explanations: The survey identifies four key features needed for effective explanations: clinical validation, consistency assessment, objective and standardized quality evaluation, and human-centered quality assessment—suggesting that no single evaluation method is sufficient to ensure explanations are actually useful to healthcare professionals and patients.

Current XAI methods have significant limitations: The paper emphasizes "main research challenges in the field as well as the limitations of existing XAI methods," indicating that practitioners should be cautious about over-relying on current explainability techniques without understanding their constraints in healthcare applications.