Detection and Semantic Segmentation of Rib Fractures using a Convolutional Neural Network Approach

Rib Fractures are one of the most common bone injuries that can occur. Over 3 million cases are diagnosed in the United States alone yearly making it a very common fracture. Causes are usually from chest trauma such as falls and sports accidents. The detection of Rib Fractures is a common task in clinics and a labor-intensive such that a specialist radiologist is required to detect rib fractures. The proposed solution looks at processing CT Scans containing rib fractures with an Attention U-Net CNN Architecture. The system achieved a segmentation dice of 77.46% and an IoU of 63.21% with an improvement of 5.96% and 7.61% respectively from the currently published paper.