SCRFD: Spatial Coherence Based Rib Fracture Detection

Rib fracture is a very common type of chest injury. Currently, the diagnosis of rib fracture is usually performed by chest CT. Due to the large number of chest CT slices, the diagnosis is very time- consuming. Especially for the detection of non-displaced fractures with very fine fracture locations and multiple fractures in the same patient's ribs, the missed diagnosis rate is still very high. To improve the detection precision, we propose a spatial coherence based rib fracture detection method, first we employ a novel feature extraction method to extract the rib region from CT slice, followed by a novel spatial coherence based convolutional neural network to recognize whether fracture occurs in the rib region. We have compared our method with currently popular object detection method. Experiment results show that our method has drastically improvements on precision and performance over previous methods.