Towards the Use of Saliency Maps for Explaining Low-Quality Electrocardiograms to End Users

Real-time quality feedback reduces patient burden: The paper addresses a practical problem in telemedicine where low-quality electrocardiograms (ECGs) are only detected after patients leave clinics, forcing them to return for repeat exams. This is particularly burdensome for patients in remote regions, demonstrating how AI explanations can improve healthcare accessibility and efficiency.

Explainability matters for non-expert users: The research includes interviews with clinic technicians to understand their specific explanation needs, recognizing that end-users of AI systems "do not have AI-specific expertise." This highlights that effective AI explanations must be tailored to the actual users' backgrounds and roles, not just designed for technical audiences.

Saliency maps show promise for visual explanations: The paper proposes using saliency maps (visual highlighting of important image regions) to explain why an ECG is flagged as low-quality, suggesting that visual explanations may be more intuitive for medical technicians interpreting medical images compared to text-based or numerical explanations.

Longitudinal evaluation is essential for responsible AI deployment: The authors note this would be "the first longitudinal study on evaluating the effects of XAI methods on end-users," emphasizing that responsible AI implementation requires studying real-world impacts over time rather than just testing systems in controlled settings.

Stakeholder input shapes AI system design: By conducting interviews with clinic staff before full deployment, the research demonstrates best practices in responsible AI—involving actual users in understanding what explanations they need to do their jobs effectively and safely.