The Detection and Classification of Scaphoid Fractures in Radiograph by Using a Convolutional Neural Network.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Scaphoid fractures are often occult and difficult to detect with standard X-rays due to subtle presentations.
• Traditional radiography has limitations in visualizing non-displaced and occult scaphoid fractures accurately.

Purpose of the Study:

• To develop and evaluate a two-stage Convolutional Neural Network (CNN) approach for enhanced scaphoid fracture detection and classification.
• To improve diagnostic accuracy for subtle scaphoid fractures using multi-view X-ray imaging.

Main Methods:

• A two-stage CNN model was implemented, utilizing Faster RCNN with FPN for scaphoid bone detection (Stage 1).
• Stage 2 involved fracture classification using a ResNet backbone, FPN, and a multi-view fusion module integrating anterior-posterior (AP) and lateral (LA) X-ray views.
• The model was evaluated on its ability to detect scaphoid bones and classify fractures, with a focus on multi-view fusion benefits.

Main Results:

• The scaphoid bone detection stage achieved 100% accuracy with high Intersection over Union (IoU) scores for both AP and LA views.
• The multi-view fusion model demonstrated improved fracture classification accuracy (89.94%), recall (87.33%), and precision (90.36%).
• Multi-view fusion enhanced fracture detection accuracy to 87.16% (AP) and 83.83% (LA) compared to single-view methods.

Conclusions:

• The proposed multi-view fusion CNN model effectively enhances the detection and classification of scaphoid fractures, especially occult and non-displaced types.
• This automated approach offers a reliable tool to assist clinicians in more efficient and accurate scaphoid fracture diagnosis.
• Multi-view imaging integration is crucial for improving the sensitivity of AI models in diagnosing subtle bone injuries.