Towards Transparent Healthcare: Local Explanation Methods

2024

1. Towards Transparent Healthcare: Advancing Local Explanation Methods in Explainable Artificial Intelligence

   Carlo Metta, Andrea Beretta, Roberto Pellungrini, Salvatore Rinzivillo, and Fosca Giannotti

   Bioengineering, Apr 2024

   RESEARCH LINE

   1

   Abs Bib HTML

   Local explanation methods, such as SHAP and LIME, are increasingly adopted to justify predictions of clinical decision support systems. However, their reliability and clinical usefulness remain limited by instability, lack of contextualization, and poor alignment with medical reasoning. In this work, we propose an enhanced pipeline for generating trustworthy local explanations in healthcare. Our approach incorporates domain constraints, medical ontologies, and temporal reasoning over patient histories. We evaluate the method on multiple clinical prediction tasks and compare it against standard explainability tools using expert-driven criteria. Results show that explanations become more stable and more aligned with clinically plausible factors. A qualitative analysis with clinicians further indicates improved interpretability and actionability, supporting safer and more transparent AI-assisted healthcare.

   @article{MBP2024b,
     author = {Metta, Carlo and Beretta, Andrea and Pellungrini, Roberto and Rinzivillo, Salvatore and Giannotti, Fosca},
     doi = {10.3390/bioengineering11040369},
     issn = {2306-5354},
     journal = {Bioengineering},
     line = {1},
     month = apr,
     number = {4},
     open_access = {Gold},
     pages = {369},
     publisher = {MDPI AG},
     title = {Towards Transparent Healthcare: Advancing Local Explanation Methods in Explainable Artificial Intelligence},
     visible_on_website = {YES},
     volume = {11},
     year = {2024}
   }