“Explaining AI Medical Models in My Way”: An LLM-Enhanced Personalized Report for Clinicians

Abstract

Clinicians are increasingly focusing on Explainable AI (XAI) to better understand AI medical models and improve decision-making. While algorithm-focused approaches dominate, the role of user characteristics—crucial for how explanations are perceived—is critical but often neglected. To address clinicians' diverse explainability needs, we developed a large language model (LLM) powered system that “translates” medical AI outputs into adaptive, characteristics-driven reports. Through a formative study with Parkinson's Disease (PD) clinicians (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{N} \boldsymbol{=} \mathbf{8}$</tex>) and prior studies, we identified that clinicians' preferences for reports, containing information Type, content Tone, and reading Time (3T), vary based on their expertise and workflow urgency. Leveraging these insights, we developed PD Report, an adaptive system powered by large language models (LLMs) to generate personalized, context-aware diagnostic explanations tailored to clinicians' expertise and workflows. A user study (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{N}=12$</tex>) validated its usability, with clinicians reporting high satisfaction, trust, and efficiency. The system achieved a System Usability Scale (SUS) score of 82.9 (excellent usability), a strong intention-to-use rating (4.8/5), and low cognitive load (NASA Task Load Index). This work highlights LLMs' potential to deliver personalized, understandable outputs from AI medical models.

Ähnliche Arbeiten