Detection of Rib Fractures on Computed Tomography Images Using Deep Learning Methods

Rib fractures are an important task in determining the severity of a thoracic injury. In order to qualitatively determine the injury, experienced radiologists are needed, moreover, the quality also depends on the doctor's skill and the number of computed tomography (CT) images that need to be processed. Many studies on this problem in one way or another aim to create an algorithm that is better able to detect and segment rib fractures. In this research, I study a deep learning model called FracNet, developed for the task of rib fracture detection and segmentation on CT images, to recreate the results of the authors of [1]. The RibFrac dataset [17] included 7,473 annotated traumatic rib fractures from 900 human-labeled patients. I entered the RibFrac 2020 [18] rib fracture analysis competition. I ranked 342nd in this competition. My results were as follows: Dice = 0.12, IoU = 0.06 (Intersection over union), average recall was 0.07, and maximum recall was 0.22.