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

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Wrist fractures (WFs) are common musculoskeletal injuries requiring accurate and timely diagnosis.
• X-ray imaging is standard for WF detection, but current deep learning models face challenges with image quality and feature focus.
• Existing methods often rely on convolutional neural networks or You Only Look Once models.

Purpose of the Study:

• To develop and evaluate a novel hybrid deep learning model for automated wrist fracture detection.
• To improve diagnostic performance by integrating local and global feature representations.
• To address limitations of current models in handling imaging variability and focusing on informative regions.

Main Methods:

• A hybrid EffViT-B6 model was proposed, combining EfficientNet-B6 and Vision Transformers.
• The model automatically extracts and integrates local and global features in a single stage.
• Gradient-weighted class activation maps were used for fracture region identification without manual annotation.

Main Results:

• The EffViT-B6 model achieved 89.07% classification accuracy and 93.21% AUC on the Musculoskeletal Radiographs (MURA) dataset.
• It demonstrated superior performance compared to existing methods on publicly available wrist X-ray datasets.
• The model effectively identified fracture regions, showcasing enhanced diagnostic capabilities.

Conclusions:

• The hybrid EffViT-B6 model offers a significant advancement in automated wrist fracture detection.
• This AI-driven approach can assist clinicians by improving diagnostic accuracy, efficiency, and timeliness.
• The model's ability to integrate multi-level features and identify fracture regions holds promise for clinical application.