Exploiting Data Augmentation Strategies to Improve the Classification of Spinal Disorders in X-Ray Images

Spinal disorders affect a significant portion of the population and are a growing concern for healthcare authorities due to their potentially debilitating consequences. In this context, computer vision techniques offer a promising path for rapid diagnosis, but they often require large datasets to train robust and reliable models. However, public datasets that can support the training of such models are scarce. To address this, we explored the application of advanced data augmentation strategies, namely CutMix, CutOut, and MixUp, combined with standard augmentation techniques to improve the performance of deep learning models in classifying X-ray images into three categories: (i) healthy, (ii) scoliosis, and (iii) spondylolisthesis. We applied these techniques by training the ResNet-50, Vision Transformer (ViT), and Swin Transformer V2 architectures, evaluating their effectiveness for this task. Our experiments revealed that the combination of ViT architecture and CutMix augmentation achieved the highest accuracy, with a performance of 0.9882.