The paper presents a comprehensive computer-aided diagnosis (CAD) system designed for early detection of knee Osteoarthritis (OA) utilizing knee medical imaging and machine learning algorithms. Osteoarthritis is a prevalent chronic disease affecting various joints, primarily the fingers, thumbs, spine, hips, knees, and big toes, with secondary occurrences linked to pre-existing joint anomalies. Although more common among older individuals, OA can develop in adults of any age, characterized by degenerative changes in joints. Traditional diagnosis involves examining joint scans, typically through X-ray analyses being conducted by trained radiologists and orthopaedists, which can be time-consuming and subject to precision loss due to manual segmentation. Automatic segmentation and interpretation of joint X-ray scans are thus necessary to enhance clinical outcomes and bone calculation precision. The advent of deep learning technologies in medical systems has facilitated such transition, enabling efficient processing of large data volumes with improved accuracy. In particular, Convolutional Neural Networks (CNNs) being among the deep learning methods, have proven effectiveness in automating X-ray scan segmentation. The paper provides an overview of various deep learning and image processing techniques employed for automatic segmentation and interpretation of X-ray scans, facilitating disease diagnosis based on image data along with a proposed improved model of Visual Geometry Group VGG-16 with edge detection using X-ray images.  A classification algorithm based on CNN and image edge detection is proposed demonstrating promising results, achieving predictive accuracies exceeding 90% across all suggested models. Particularly is the performance of the proposed VGG-16 after training with edge detection, which attained a training accuracy of 100% and a testing accuracy of 98.2%. This highlights the efficacy of deep learning approaches in enhancing diagnostic accuracy and efficiency in knee OA detection. A comparative evaluation of the proposed algorithm against other techniques based on performance metrics is reported.