Group-wise Compression and Summarization via LLM-based Ensemble for Chest X-ray Report Generation

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Natural Language Processing for Clinical Documentation

Background:

• Chest X-ray reports are crucial for diagnosing medical conditions but manual creation is time-consuming.
• Accurate interpretation and clinical consistency are vital in radiology report generation.
• Automating report generation can significantly aid radiologists and improve efficiency.

Purpose of the Study:

• To develop and evaluate a novel framework for automated chest X-ray report generation.
• To enhance the accuracy, reliability, and clinical relevance of AI-generated radiology reports.
• To reduce the manual workload for radiologists in creating chest X-ray reports.

Main Methods:

• A two-step Large Language Model (LLM) ensemble approach combined with disease-based retrieval.
• Image-text embedding space similarity for initial report retrieval.
• Filtering retrieved reports using patient-specific disease information for clinical relevance.
• Progressive summarization by the LLM ensemble to refine reports and preserve diagnostic insights.

Main Results:

• The proposed framework demonstrated superior performance on MIMIC-CXR and IU X-ray benchmarks.
• Achieved improved clinical relevance and diagnostic reliability in automated report generation.
• Effectively reduced redundancy and enhanced coherence in the generated reports.
• Showcased significant potential in reducing radiologist workload while maintaining report quality.

Conclusions:

• The novel LLM ensemble with disease-based retrieval offers a promising solution for automated chest X-ray report generation.
• This approach enhances diagnostic reliability and clinical relevance, aiding medical professionals.
• The framework represents a significant advancement in AI-assisted medical documentation.