Composite Attention Residual U-Net for Rib Fracture Detection

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Computed tomography (CT) is crucial for diagnosing rib fractures and chest trauma severity.
• Manual identification of rib fractures in numerous CT scans is time-consuming and challenging for radiologists.

Purpose of the Study:

• To develop an automated, precise, and rapid method for detecting rib fractures using CT images.
• To improve the diagnostic accuracy and efficiency of radiologists in identifying rib fractures.

Main Methods:

• A U-net-based deep learning model was developed for pixel-level rib fracture feature extraction.
• The model incorporates a Combined Attention Module (CAM) for feature fusion and a Hybrid Dense Dilated Convolution Module (HDDC) for semantic information acquisition.

Main Results:

• The model achieved a Recall of 81.71%, F1 score of 81.86%, and Dice coefficient of 53.28% on a public dataset.
• The proposed method demonstrated higher detection sensitivities compared to human-only or computer-only diagnoses, assisting radiologists in reducing false positives.

Conclusions:

• The U-net-based model effectively detects rib fractures in CT images, enhancing diagnostic capabilities.
• Integrating this AI tool can significantly improve radiologists' ability to diagnose chest trauma with greater speed and accuracy.