Segmentation and Classification of Joint Space Narrowing in Rheumatoid Arthritis Using Deep Learning

Area of Science:

• Radiology
• Artificial Intelligence
• Rheumatology

Background:

• Rheumatoid arthritis (RA) is a chronic autoimmune disease.
• Joint space narrowing (JSN) is a key diagnostic marker for RA.
• Accurate JSN assessment is crucial for patient management.

Purpose of the Study:

• To develop and evaluate a deep learning approach for classifying JSN severity using X-ray images.
• To compare the performance of different image segmentation techniques for JSN analysis.
• To introduce a novel masking method, Halo Edge Mask (HEM), for improved bone segmentation.

Main Methods:

• Utilized deep learning for JSN severity classification on X-ray images.
• Employed YOLOv5 for joint region segmentation.
• Applied a fine-tuned Segment Anything Model (SAM) for bone mask generation.
• Introduced and compared the Halo Edge Mask (HEM) with other masking techniques.

Main Results:

• The proposed HEM method demonstrated superior performance across all evaluated metrics.
• HEM achieved higher sensitivity, specificity, and AUROC compared to other masking methods.
• The AI-based approach showed feasibility for JSN assessment.

Conclusions:

• AI-assisted assessment of JSN severity using X-rays is feasible.
• The HEM method enhances the accuracy of AI-driven musculoskeletal disorder analysis.
• This technology can support clinical decision-making, especially in resource-limited settings.