Rib Fracture Detection with Dual-Attention Enhanced U-Net.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Rib fractures are frequent consequences of chest trauma, necessitating accurate diagnosis.
• Current deep learning methods, like convolutional neural networks (CNNs), struggle with rib fracture detection due to data limitations and fracture complexity.
• Existing CNN-based approaches exhibit suboptimal accuracy and sensitivity in identifying rib fractures.

Purpose of the Study:

• To develop an advanced deep learning model for improved rib fracture detection.
• To enhance the feature extraction capabilities of CNNs for irregular fracture shapes.
• To increase the accuracy and sensitivity of automated rib fracture diagnosis.

Main Methods:

• Proposed the CFSG U-Net, integrating a U-Net architecture with a dual-attention module.
• Implemented a channel-wise fusion attention module (CFAM) to refine feature maps.
• Incorporated a spatial-wise group attention module (SGAM) for capturing fine-grained spatial information.
• Utilized a newly established rib fracture dataset for model evaluation.

Main Results:

• The CFSG U-Net achieved a maximum sensitivity of 89.58%.
• The proposed method obtained an average FROC score of 81.28%.
• CFSG U-Net demonstrated superior performance compared to existing detection methods and attention modules.

Conclusions:

• The CFSG U-Net effectively addresses the limitations of previous CNN-based methods for rib fracture detection.
• The dual-attention mechanism significantly improves the model's ability to detect complex rib fractures.
• This approach offers a promising advancement for accurate and sensitive automated diagnosis of rib fractures.