Comparative Analysis of Computer Vision Models for Image Classification on Wrist X-rays in the MURA Dataset

Selecting the optimal deep learning model for medical image classification, particularly for wrist X-rays, requires balancing accuracy, computational efficiency and model complexity. In this study, we compare six prominent architectures-YOLOv8, ResNet, Vision Transformers, VGGNet, EfficientNet, and Inception Networks on the MURA dataset to assess their performance in terms of classification accuracy and computational efficiency for a subset of wrist x-rays. Experimental results show that YOLOv8 achieves the highest accuracy (87%), followed by ViT (84%), ResNet (83%), EfficientNet (82%), VGGNet(81%) and Inception Networks (76%). YOLOv8’s superior feature extraction capabilities make it the best choice for small-scale medical imaging tasks, offering valuable insights for model selection in clinical applications. These findings emphasize the importance of considering both accuracy and efficiency when selecting deep learning models for medical image analysis.