A 2-step deep learning approach to splenic injury detection

The spleen is the most commonly injured solid organ in blunt abdominal trauma. Physicians can use abdominal CT to understand the location of the injury and determine its severity grade. However, these injuries may go undetected because of the inexperience of junior physicians. In our previous study, we had performed a 2-step object detection based analysis pipeline on abdominal CT images to identify whether the spleens are injured or not, and also reported that localizing the spleen first can enhance the prediction performance. In order to delineate the region of the spleen more precisely, in this study we adopted TotalSegmentator without the need of the training process to segment the contour of the spleen. In addition, we further utilized morphology operations to expand the segmented region, especially useful for the spleen injury cases. Our proposed method improves the baseline which only using TotalSegmentator to get the segmentation of spleen by 6% in terms of accuracy, which confirms its effectiveness.