.png){kind=icon}
.png){kind=icon}
Return
Experience-guided multi-agent interpretable framework for radiology report summarization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.cmpb.2025.109078.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• This paper introduces EMAI, an innovative framework for radiology report summarization that integrates a self-evolutive nearest-neighbor explicit experience induction algorithm, an interpretable findings analysis module, and a collaborative multi-agent architecture. The experience induction algorithm automatically extracts and generalizes knowledge from historical reports, while the findings analysis module deconstructs and explains the relationships between findings and impressions. The multi-agent framework adaptively determines when and how to leverage experiences or relevant reports for impression generation, dynamically integrating the components to enhance the accuracy and interpretability of the generated impressions. • EMAI Demonstrates significant improvements in the accuracy and interpretability of impression predictions on two public datasets, MIMIC-CXR and Open-I, compared to state-of-the-art baselines. The induction learning process enables large language models to generalize better from diverse case data, while the interpretability mechanism provides valuable explanations that enhance the trustworthiness of the generated results. The experimental results showcase the effectiveness of the proposed framework in generating accurate and interpretable impressions, highlighting its potential to assist radiologists in diagnosis and treatment planning. • Highlights the potential of integrating large language models with agent-based frameworks to improve clinical decision-making and advance the efficiency of radiology workflow. The EMAI framework has shown the strengths of large language models in capturing diagnostic knowledge and generating human-readable explanations, while the multi-agent architecture allows for dynamic integration of experiences and relevant reports. This result not only improves the quality of radiology report summarization but also provides clinically grounded information that can assist radiologists in their daily practice.
Journal
IF:
4.8
Papers:
6.9K
Citations:
2.1W
Organization
View more organizations