CNN-ELM-BASED DEEP LEARNING FRAMEWORK FOR KNEE OSTEOARTHRITIS CLASSIFICATION FROM RADIOGRAPHIC IMAGES

• Published in: International journal of advances in signal and image sciences Vol. 11; no. 1; pp. 17 - 29
• Main Authors: Srividhya, V, Juliet, P Jega, Neelima, N, Seeni, Senthil Kumar
• Format: Journal Article
• Language: English
• Published: XLESCIENCE 30.06.2025
• Subjects: knee osteoarthritis, deep learning, convolutional neural network, extreme learning machine, medical image processing, radiograph analysis
• ISSN: 2457-0370; 2457-0370
• DOI: 10.29284/IJASIS.11.1.2025.17-29

This paper proposes a deep learning framework for automated Knee OsteoArthritis (KOA) severity classification from radiographic images, using a hybrid custom Convolutional Neural Network and Extreme Learning Machine (CNN-ELM) architecture. The CNN-ELM model system integrates Contrast Limited Adaptive Histogram Equalisation (CLAHE) in the preprocessing stage to enhance image quality and highlight subtle structural differences associated with KOA. The custom CNN composed of three convolutional layers extracts deep spatial features from the enhanced X-ray images and these features are passed to the ELM classifier, which performs fast, non-iterative learning using pseudo-inverse computations. Experimental evaluation on a public KOA X-ray dataset comprising 9,786 images demonstrates significant improvements in classification performance, particularly for severe KOA cases. The proposed system achieves an average accuracy of 99.13% and an F1-score of 98.93% with preprocessing, outperforming standard CNN models and existing baseline methods such as ResNet34, VGG16, and DenseNet. These results highlight the effectiveness of the CNN-ELM model and preprocessing pipeline in enhancing the precision and reliability of KOA severity diagnosis.