A Research on Result Interpretability of Medical AI Based on Large Language Model
Keywords: XAI, AI Agent, Medical AI, Amyloidosis, LLM, AIGC
Abstract: Explainability is one of the important challenges facing the application of medical AI. The existing AI explainability research is more of a kind of process explainability study. Drawing on the behavioral habits of human beings to communicate on a certain topic, this paper proposes a definition of result interpretability for medical AI, divides explainable medical AI research into three phases: data explainability, process explainability and result interpretability, and argues that once an AI model reaches a certain result interpretability metric, we can accept its conclusions and apply it to the clinic without having to wait until human beings fully understand the operation and decision-making mechanism of the AI model before using it. In this regard, we propose the c oncept of interpretative integrity. Further, we propose an architecture for result-interpretable medical AI system based on AI-Agent and build a result-interpretable system around risk prediction AI model for amyloidosis, which enables professional interpretation of the result of the risk prediction model for amyloidosis disease through a large language model and supports professional Q&A with clinicians. The implementation of the system enhances clinicians' professional acceptance of medical AI models, and provides a more feasible realization path for the large-scale application of AI-assisted diagnosis.
Primary Area: interpretability and explainable AI
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1525
Loading