Rib Fracture Detection with Dual-Attention Enhanced U-Net

Abstract

Rib fractures are common injuries caused by chest trauma, which may cause serious consequences. It is essential to diagnose rib fractures accurately. Low-dose thoracic computed tomography (CT) is commonly used for rib fracture diagnosis, and convolutional neural network- (CNN-) based methods have assisted doctors in rib fracture diagnosis in recent years. However, due to the lack of rib fracture data and the irregular, various shape of rib fractures, it is difficult for CNN-based methods to extract rib fracture features. As a result, they cannot achieve satisfying results in terms of accuracy and sensitivity in detecting rib fractures. Inspired by the attention mechanism, we proposed the CFSG U-Net for rib fracture detection. The CSFG U-Net uses the U-Net architecture and is enhanced by a dual-attention module, including a channel-wise fusion attention module (CFAM) and a spatial-wise group attention module (SGAM). CFAM uses the channel attention mechanism to reweight the feature map along the channel dimension and refine the U-Net’s skip connections. SGAM uses the group technique to generate spatial attention to adjust feature maps in the spatial dimension, which allows the spatial attention module to capture more fine-grained semantic information. To evaluate the effectiveness of our proposed methods, we established a rib fracture dataset in our research. The experimental results on our dataset show that the maximum sensitivity of our proposed method is 89.58%, and the average FROC score is 81.28%, which outperforms the existing rib fracture detection methods and attention modules.

FasterRib: A deep learning algorithm to automate identification and characterization of rib fractures on chest computed tomography scans

“…To our knowledge, six prior studies used the RibFrac or other private institutional CT scan data sets to build rib fracture prediction algorithms. Five used U-Net or V-Net-based models; 5,7–11 these models approached fracture detection as a segmentation task, which lacks computational efficiency compared with object detection tasks. One study used a Faster R-CNN–based object detection model, 10 but FasterRib used a larger ResNet backbone and deeper classifier architecture to improve detection performance.…”

“…Existing rib fracture detection algorithms posed detection as a segmentation task, wherein every pixel per CT scan slice require binarily labeling (1 for fracture, 0 otherwise). 5,[7][8][9][10][11] To mitigate computational cost, we reformulated the problem into one of object detection, whereby the model outputs bounding boxes around each fracture per slice (boxes formed by pixel coordinates of the upper left and bottom right extremes per predicted fracture).…”

Diagnostic Performance of Artificial Intelligence in Rib Fracture Detection: Systematic Review and Meta-Analysis

“…On the remaining 32 studies, a full-text assessment was performed, through which 20 more studies were excluded. In the end, 12 studies were included in the systematic review [17][18][19][20][21][22][23][24][25][26][27][28], all of which were identified in the initial search. Unpublished relevant studies were not obtained.…”

“…The reference standard was quite similar for all the included studies, and all of them were retrospective. The quality of the studies ranged from intermediate to high quality: 2 studies with intermediate quality [18,19] and 10 studies with high quality [17,[20][21][22][23][24][25][26][27][28].…”

“…The reference standard was quite similar for all the included studies, and all of them were retrospective. The quality of the studies ranged from intermediate to high quality: 2 studies with intermediate quality [18,19] and 10 studies with high quality [17,[20][21][22][23][24][25][26][27][28]. The number of included patients and included CT scans was deemed as being the same, as it is likely that the large majority of included patients underwent a single CT scan.…”