Automated Diagnosis of Rheumatoid Arthritis From Hand Radiographs Using Artificial Intelligence: A Retrospective...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Rheumatology

Background:

• Rheumatoid arthritis (RA) is a chronic inflammatory condition affecting hand and wrist joints, causing pain and disability.
• Current radiographic assessment for RA diagnosis is subjective and relies heavily on clinician expertise.
• Deep learning offers potential for objective, consistent, and efficient evaluation of radiographic changes.

Purpose of the Study:

• To develop an attention-based deep learning model for automated rheumatoid arthritis diagnosis using hand and wrist radiographs.
• To evaluate the model's diagnostic performance, particularly with limited datasets.
• To demonstrate the potential for objective and consistent RA assessment.

Main Methods:

• A retrospective observational study analyzed radiographs from 311 RA patients and 259 controls.
• DenseNet121 and DenseNet169 architectures were combined with an attention mechanism.
• Data augmentation and attention modeling were employed to enhance robustness on a limited dataset.

Main Results:

• The attention-based deep learning model achieved 88% accuracy, 84% precision, and 91% recall.
• The model demonstrated strong diagnostic capability despite the limited training data.
• Initial clinical testing indicated the model can assist radiologists with consistent, objective assessments.

Conclusions:

• An attention-based deep learning approach shows promise for effective, reliable, and efficient automated RA diagnosis.
• High performance with limited data suggests potential for clinical adoption, especially in resource-limited settings.
• The model can serve as a valuable tool to support radiologists in diagnosing rheumatoid arthritis.