A K-nearest neighbors-based classification approach for automated detection of knee osteoarthritis

Purpose: Osteoarthritis is a serious condition that can significantly reduce a person's quality of life, causing pain and stiffness in the knees and limiting their mobility. The condition progressively worsens over time, emphasizing the importance of early diagnosis. This study implemented a computer-aided classification approach to reduce the time and effort required for diagnosing knee osteoarthritis while minimizing human errors. Materials and Methods: Data analyzed in this study was obtained from the Osteoarthritis Initiative. A total of 165 samples were used in the study. All abnormal samples were graded as severe osteoarthritis. While 78 samples were used to test the implemented algorithm, the training process of the algorithm was completed with 87 samples. The proposed approach involves three main stages: segmenting the cartilage region through a series of image -processing operations, extracting morphological features from the defined region, and classifying samples based on these features. In the classification stage, morphological features characterizing the cartilage region were classified in the observation space, and the k-nearest neighbors algorithm was applied for automated discrimination. Accordingly, the computer utilizes the previously classified sample features to estimate the presence of pathology. Results: Test classifications were completed with 78 samples; 28 were previously diagnosed with osteoarthritis. Morphological measures of the training samples were accepted as a reference for abnormality. The applied classification scheme can distinguish severed cartilage regions with a 0.95% accuracy. Conclusion: This study demonstrates the potential effectiveness of a computer-aided approach in diagnosing knee osteoarthritis with high accuracy. The developed approach offers a promising solution for early and efficient diagnosis, enabling more timely and effective treatment strategies for osteoarthritis patients. The progressive nature of the disease makes these advancements in diagnostic methods invaluable. Future studies may focus on expanding the sample size and further refining the model for enhanced precision and broad applicability in clinical settings.