An Efficient Arthritis Diagnostic Model using CoAtNet by Integrating Convolutional and Attention Mechanisms for Enhanced Precision

Osteoarthritis (OA) is a common joint disease that lowers quality of life and causes damage worldwide. It is more common in elderly people, with knee OA being the most common type. While traditional machine learning and deep learning methods have validated potential in OA identification, they normally struggle to balance spatial feature abstraction and interpretability. Interestingly, the effective combination of convolutional and attention mechanisms to enhance diagnosis accuracy, specifically in OA severity categorization, remains unexplored. Our goal is to minimize this performance discrepancy by developing an efficient diagnostic model for knee OA utilizing CoAtNet. This hybrid architecture combines the advantages of attention mechanisms with convolutional layers for optimal performance. The Kaggle knee osteoarthritis dataset was used to train and assess the model thoroughly. To ensure robustness, key preprocessing methods such as image normalization and data augmentation were used. Experimental results show that the suggested CoAtNet-based model obtains $\mathbf{9 6 \%}$ accuracy, 88% F1 scores, and 87% precision, culminating in an astounding Top 1 accuracy of $\mathbf{9 6 \%}$. In addition to providing a more precise means of diagnosing OA, this model offers a scalable instrument that may be easily included in clinical processes. Using this method could improve knee OA early identification and treatment planning, which would improve patient results and boost overall healthcare proficiency.