Explainable Artificial Intelligence For The Detection and Characterisation of Stage B Heart Failure

XAI methods exist but lack consistency: While the review identified SHAP and LIME as commonly used explainability techniques for detecting early-stage heart failure, adoption was inconsistent across studies, with some relying on "limited or ad hoc interpretability approaches," suggesting the field has not yet standardized how AI systems explain their decisions in this clinical context.

Critical gap in fairness evaluation: None of the 20 studies examined whether AI explanations differed across sex or ethnic subgroups, despite evidence that heart failure affects different populations differently. This represents a significant oversight in ensuring the technology works equitably for all patients.

Explanations are often not rigorously validated: The review found that many studies either failed to assess their explanations at all or only compared them against published literature, which can introduce bias. This means clinicians cannot be confident that the AI's reasoning is actually sound or trustworthy.

Early detection potential requires responsible implementation: While XAI could help identify asymptomatic heart failure patients early enough for preventive treatment, realizing this benefit depends on addressing the current gaps in validation, fairness testing, and transparent reporting across demographic groups.

Explainability alone is insufficient without systematic evaluation: The paper demonstrates that simply using explainability techniques is not enough—they must be rigorously tested and compared across different patient populations to ensure AI recommendations are both accurate and fair for clinical use.