Automated fracture screening using an object detection algorithm on whole-body trauma computed tomography

Abstract

The emergency department is an environment with a potential risk for diagnostic errors during trauma care, particularly for fractures. Convolutional neural network (CNN) deep learning methods are now widely used in medicine because they improve diagnostic accuracy, decrease misinterpretation, and improve efficiency. In this study, we investigated whether automatic localization and classification using CNN could be applied to pelvic, rib, and spine fractures. We also examined whether this fracture detection algorithm could help physicians in fracture diagnosis. A total of 7664 whole-body CT axial slices (chest, abdomen, pelvis) from 200 patients were used. Sensitivity, precision, and F1-score were calculated to evaluate the performance of the CNN model. For the grouped mean values for pelvic, spine, or rib fractures, the sensitivity was 0.786, precision was 0.648, and F1-score was 0.711. Moreover, with CNN model assistance, surgeons showed improved sensitivity for detecting fractures and the time of reading and interpreting CT scans was reduced, especially for less experienced orthopedic surgeons. Application of the CNN model may lead to reductions in missed fractures from whole-body CT images and to faster workflows and improved patient care through efficient diagnosis in polytrauma patients.

Figure 1

Flow chart showing the overall study process. CT, computed tomography.

Figure 2

A screenshot of the image annotation process. We removed the black background and trimmed to include the area of the patient's trunk. A rectangular bounding box covering the minimum area of the fracture site was then drawn on every fracture of the CT slice and was labeled as spine fractures, pelvic fractures, or rib fractures using labelImg (version: 1.8.1, available at https://github.com/tzutalin/labelImg). CT, computed tomography.

Figure 3

Sample CT images correctly annotated by neural networks on the testing dataset. (A and D) Pelvic fracture. (B and E) Rib fracture. (C and F) Spine fracture. The CNN model also generated a confidence score for each of the detected points as continuous values of a range from 0 to 100%). CT, computed tomography; CNN, convolutional neural network.

Figure 4

Representative images of true positive, false positive, and false negative. (A–C) CT axial slices detected and diagnosed correctly by the CNN model for three fractures (true positive). (D–F) CT axial slices of a misdiagnosed fracture (false positive). The vascular groove was mistaken for a pelvic fracture (D); The boundary between costal cartilage and rib was mistaken for a rib fracture (E); The vascular groove was mistaken for a spine fracture (F). (G–I) CT axial slices of a missed fracture (false negative were indicated by arrowheads). CT, computed tomography; CNN, convolutional neural network.