Self-Reflective Chest X-Ray Report Generation with Clinical-Aware Detection and Multilevel Readability

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Natural Language Processing

Background:

• Clinical documentation requires balancing diagnostic precision with patient understanding.
• Automated solutions are needed to adapt medical report complexity to varying literacy levels.
• Current methods often struggle to maintain clinical rigor while improving patient accessibility.

Purpose of the Study:

• To develop and validate a unified framework for automated medical documentation.
• To dynamically adapt report complexity to diverse literacy levels (6th, 11th, 18th grade).
• To establish evaluation methodologies for patient-centered medical documentation.

Main Methods:

• Integrated a hybrid detection method (CheXFusion, Eigen-CAM) for finding detection and localization.
• Utilized a LLaVA-based pipeline for synthesizing clinical predictions with anatomical data.
• Employed a self-reflective large language model for dynamic complexity adaptation and novel evaluation metrics (Mistral-small).

Main Results:

• Achieved a 19.78% enhancement in classification accuracy (AUROC) and 17.29% improvement in mean average precision.
• Demonstrated a 56.88% increase in patient comprehension scores and a 5.26% gain in diagnostic precision.
• Successfully balanced clinical rigor with enhanced patient accessibility.

Conclusions:

• The developed framework sets new standards for automated medical documentation.
• It effectively reconciles clinical precision with patient comprehension in healthcare communication.
• Reduces provider burden and improves patient engagement through clear, accessible reports.