Evidence-Guided Diagnostic Reasoning for Pediatric Chest Radiology Based on Multimodal Large Language Models

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatric Pulmonology

Background:

• Pediatric respiratory diseases are a major cause of hospitalization and mortality.
• Chest radiography is the primary imaging tool for pediatric lung assessment.
• Current AI diagnostic tools lack interpretability and clinical trustworthiness.

Purpose of the Study:

• To develop a trustworthy, two-stage AI diagnostic paradigm for pediatric chest X-rays.
• To enhance diagnostic accuracy and clinical workflow alignment.
• To improve the interpretability of AI in pediatric radiology.

Main Methods:

• A two-stage approach: 1) A vision-language model identifies radiological findings from X-rays.
• 2) A multimodal large language model integrates findings, patient data, and medical knowledge (RAG).
• Utilized the VinDr-PCXR dataset for model training and evaluation.

Main Results:

• Achieved 90.1% diagnostic accuracy, 70.9% F1-score, and 82.5% AUC.
• Demonstrated up to a 13.1% increase in diagnostic accuracy compared to existing methods.
• Validated the effectiveness of multimodal reasoning with explicit evidence and domain knowledge.

Conclusions:

• The proposed AI paradigm offers trustworthy and interpretable diagnosis for pediatric chest X-rays.
• This approach shows significant potential for clinical application in pediatric radiology.
• Combining multimodal reasoning with evidence-based knowledge enhances AI diagnostic capabilities.