Potential of AI-based diagnostic grading system for knee osteoarthritis

Area of Science:

• Musculoskeletal imaging
• Artificial intelligence in healthcare
• Radiology and diagnostic imaging

Background:

• Knee osteoarthritis (KOA) is a leading cause of global disability.
• Manual radiographic grading of KOA is variable and lacks accuracy.
• AI-based computer-aided diagnostic (CAD) systems offer improved detection and grading.

Purpose of the Study:

• Develop and evaluate an AI diagnostic grading system for KOA.
• Utilize X-ray imaging and transfer learning for precise KOA diagnosis.
• Assist clinicians and trainees in early and accurate KOA detection.

Main Methods:

• Cross-sectional study with 301 knee radiographs (602 images).
• Transfer learning with DenseNet-121 as the base network.
• Performance evaluation against other deep learning and classical ML models.

Main Results:

• DenseNet-121 achieved 68.85% accuracy and 85.67% AUC.
• Comparative models showed lower performance (e.g., ≈60% accuracy).
• Dataset imbalance and difficulty distinguishing early grades were challenges.

Conclusions:

• AI-based CAD system shows potential for KOA diagnosis and grading support.
• DenseNet-121 demonstrated reliable performance despite data limitations.
• Future work includes larger datasets and multi-center validation for generalizability.