Related Experiment Video
Updated: Jul 27, 2026
02:09
Multi-modal Pulmonary Imaging: Using Complementary Information from CT and Hyperpolarized 129Xe MRI to Evaluate Lung Structure-Function
Published on: April 12, 2024
Performance evaluation of domain-specific and general-purpose AI models for chest radiograph interpretation: a
Tae-Hoon Kim1, Jun Hyung Hong1, Jihun Hyun2
1Department of Radiology, Chosun University Hospital and Chosun University College of Medicine, 365 Pilmun-Daero, Dong-Gu, Gwangju, 61453, Republic of Korea.
BMC Medical Imaging
|July 11, 2026
Summary
A specialized AI model (M4CXR) showed higher diagnostic consistency for chest X-rays than a general AI (ChatGPT-4o). This domain-specific AI can aid radiologists, improving efficiency and accuracy in medical imaging interpretation.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Radiology AI
- Machine Learning for Healthcare
Background:
- Chest radiography interpretation is challenging due to anatomical complexity and subtle findings.
- Multimodal large language models (LLMs) offer automated radiology report generation, but clinical validation against specialized AI is needed.
Purpose of the Study:
- To compare the performance of a domain-specific AI (M4CXR) against a general LLM (ChatGPT-4o) for chest radiograph interpretation.
- To evaluate clinical applicability and diagnostic consistency of AI models in radiology.
Main Methods:
- Retrospective analysis of 500 anonymized chest radiographs.
- Evaluation of AI-generated reports by four board-certified radiologists.
- Assessment of key finding detection, report generation time, and discrepancies using RADPEER scoring.
Main Results:
- M4CXR achieved significantly higher report consistency (55.8% vs. 19.8%) and lower inconsistency rates compared to GPT-4o.
- M4CXR reduced report generation time substantially (16.3s vs. 179.2s).
- Good reliability (ICC=0.701) and substantial agreement (κw=0.652) were found between original and M4CXR-assisted interpretations.
Conclusions:
- Domain-specific AI (M4CXR) demonstrated superior diagnostic consistency over a general LLM (GPT-4o) for chest radiograph interpretation.
- Specialized AI models show potential as assistive tools in radiology.
- Future research should focus on human-AI collaboration and multi-center validation.
