Knee osteoarthritis severity detection using deep inception transfer learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Osteoarthritis (OA) diagnosis relies on radiograph interpretation, which can be challenging for early stages.
• The Kellgren and Lawrence (KL) grading system is standard but interpretation accuracy varies.
• Existing AI models for OA detection show inconsistent performance.

Purpose of the Study:

• To develop an improved artificial intelligence (AI) model for accurate osteoarthritis (OA) severity classification.
• To enhance the identification of OA severity levels using a transfer learning approach.
• To improve upon existing AI models for OA diagnosis.

Main Methods:

• A transfer learning approach using an InceptionV3 (IV3) model was fine-tuned on the Osteoarthritis Initiative dataset.
• Dual-stage preprocessing and convolutional neural networks were employed for feature extraction.
• The performance of the fine-tuned IV3 (FT-IV3) model was compared against the original IV3 model.

Main Results:

• The FT-IV3 model achieved higher accuracies: 96.33% (training), 93.82% (validation), and 92.25% (testing).
• The original IV3 model achieved accuracies of 91.64% (training), 82.04% (validation), and 86.20% (testing).
• Cohen's Kappa value for FT-IV3 (90.69%) was superior to IV3 (83.15%), indicating better OA severity diagnosis.

Conclusions:

• The fine-tuned InceptionV3 (FT-IV3) model demonstrates superior performance in classifying osteoarthritis severity.
• This AI approach shows significant potential for improving the accuracy of OA diagnosis, particularly for moderate and severe grades.
• The study highlights the effectiveness of transfer learning in enhancing medical image analysis for orthopedic conditions.