Computer Model for Detecting Cervical Spine Fractures based on Computed Tomography Images

Abstract

The paper discusses methods of detecting cervical spine fractures based on computed tomography images using machine learning algorithms. Solving such a problem is important in the context of providing emergency care to patients with suspected spinal fractures, when accurate decision-making based on radiological data can be critical. In this case, such a machine learning model can speed up the work of a radiologist and reduce the importance of the human factor in making important decisions. After a review of analogs from the literature, it has been found that convolutional neural networks appear to be the most promising method. Using a publicly available dataset, a model "Fracture detection 3" based on a convolutional neural network is developed to solve the problem. The model demonstrates a classification accuracy of 98.25%, sensitivity of 99%, and specificity of 97.5%, which is ahead of the literature. For comparison with traditional methods, models based on the support vector method, decision tree, and k-nearest method are developed using a similar dataset. "Fracture detection 3" outperforms all developed models based on traditional methods in terms of classification accuracy.