Automatic Knee Osteoarthritis Diagnosis from Plain Radiographs: A Deep Learning-Based Approach

Area of Science:

• Orthopedics
• Medical Imaging
• Artificial Intelligence

Background:

• Knee osteoarthritis (OA) is a prevalent musculoskeletal disorder.
• Current OA diagnosis relies on subjective symptom assessment and radiographic evaluation.
• Limitations in current diagnostic methods necessitate objective and automated approaches.

Purpose of the Study:

• To develop and validate a transparent, computer-aided diagnosis (CADx) method for automatic knee OA severity scoring.
• To utilize a Deep Siamese Convolutional Neural Network for grading OA based on the Kellgren-Lawrence scale.
• To enhance diagnostic objectivity and clinical trust in automated OA assessment.

Main Methods:

• A Deep Siamese Convolutional Neural Network was developed for knee OA grading.
• The model was trained on data from the Multicenter Osteoarthritis Study.
• Validation was performed on 3,000 subjects (5,960 knees) from the Osteoarthritis Initiative dataset.

Main Results:

• The method achieved a quadratic Kappa coefficient of 0.83.
• Average multiclass accuracy reached 66.71% compared to expert annotations.
• Radiological OA diagnosis demonstrated an area under the ROC curve of 0.93.
• Attention maps provided transparency by highlighting key features influencing decisions.

Conclusions:

• The developed AI model offers a reliable and transparent method for automated knee OA severity scoring.
• The approach shows potential for improving clinical decision-making and advancing OA research.
• Openly releasing training codes and datasets facilitates further research and development.