A Hybrid EffViT-B6 Model for Automated Wrist Fracture Detection Using X-Ray Imaging

Wrist fractures (WFs) are prevalent musculoskeletal injuries with a high incidence rate. X-ray imaging, a cornerstone of conventional radiography, is widely employed for WF detection. Accurate fracture detection is essential for developing effective clinician support systems, particularly in emergency settings where timely diagnosis is critical. Existing literature has primarily focused on convolutional neural network-based architectures and You Only Look Once models for this task. Despite technological advancements, challenges persist, including variability in imaging quality and the inconsistent focus of models on the most informative regions, which hinders the performance of current end-to-end approaches. To address these challenges, we propose a hybrid EffViT-B6 model that combines the strengths of EfficientNet-B6 and Vision Transformers. The proposed model automatically extracts and integrates both local and global feature representations in a single stage, achieving enhanced diagnostic performance and identifying fracture regions through gradient-weighted class activation maps without the need for manual annotations or predefined region-of-interest detection. We evaluated the EffViT-B6 model using two publicly available baseline datasets, MURA and GRAZPEDWRI-DX, which include a diverse set of wrist X-ray images. Our results demonstrate that EffViT-B6 outperforms existing methods, achieving a classification accuracy improvement of 89.07 and an AUC of 93.21 on the publicly available baseline WF dataset (i.e., Musculoskeletal Radiographs), highlighting its superior classification performance.Clinical Relevance- This study presents a hybrid deep learning model for automated wrist fracture detection, which could assist clinicians by enhancing diagnostic accuracy, efficiency, and timeliness.