Automatic joint inflammation estimation based on regression neural networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Rheumatology

Background:

• Quantitative MRI assessment of inflammation in rheumatoid arthritis (RA) is vital for disease management.
• Traditional visual evaluation of MRI signs like bone marrow edema (BME), tenosynovitis, and synovitis is subjective and time-consuming.

Purpose of the Study:

• To develop an automated deep learning (DL)-based system for MRI analysis of inflammatory signs in RA.
• To facilitate more efficient and objective inflammation assessment for RA diagnosis and research.

Main Methods:

• Developed the Automatic DL-based MRI analysis of Inflammatory signs in RA (ADMIRA) system.
• Utilized DL models on 2254 subjects' MRI scans (wrist, MCP, MTP joints) with pre- and post-processing.
• Ensured robust evaluation using training, monitoring, testing, and validation sets with Pearson's and Intra-class correlation coefficients.

Main Results:

• ADMIRA achieved high performance, with mean R/ICCs near 0.9 for synovitis and tenosynovitis on test sets.
• The system demonstrated performance comparable to human experts, with slightly lower scores for BME.
• Visualization confirmed DL model inference processes align with expert knowledge.

Conclusions:

• ADMIRA offers accurate, expert-level inflammation estimation for RA, particularly for synovitis and tenosynovitis.
• The automated system provides a fast and reliable alternative to manual MRI analysis.
• ADMIRA has the potential to reduce labor costs and enhance diagnostic efficiency in rheumatology.