Figure 5
Impact of the Rib Fracture Detection System in clinical practice for patients with suspicion of rib fracture in the Department of Radiology. In a cohort of patients with suspected rib fractures who underwent chest CT investigation, radiologists should pay close attention to all of the ribs without the help of our model in order to look for 18.2% of the fractured ribs. Since 80.9% of the ribs were diagnosed as non-fracture ribs by this model with a 96.9% true-negative rate, it demonstrated high accuracy in identifying true non-fractured ribs by the constructed model. As a result, with the assistance of this deep learning model, radiologists only had to pay more attention to 19.1% of the ribs that were categorized as high-risk for fracture, which significantly reduced their workload in detecting rib fracture.
