Performance of the deep convolutional neural network based magnetic resonance image scoring algorithm for differentiating between tuberculous and pyogenic spondylitis

Abstract

The purpose of this study was to evaluate the performance of the deep convolutional neural network (DCNN) in differentiating between tuberculous and pyogenic spondylitis on magnetic resonance (MR) imaging, compared to the performance of three skilled radiologists. This clinical retrospective study used spine MR images of 80 patients with tuberculous spondylitis and 81 patients with pyogenic spondylitis that was bacteriologically and/or histologically confirmed from January 2007 to December 2016. Supervised training and validation of the DCNN classifier was performed with four-fold cross validation on a patient-level independent split. The object detection and classification model was implemented as a DCNN and was designed to calculate the deep-learning scores of individual patients to reach a conclusion. Three musculoskeletal radiologists blindly interpreted the images. The diagnostic performances of the DCNN classifier and of the three radiologists were expressed as receiver operating characteristic (ROC) curves, and the areas under the ROC curves (AUCs) were compared using a bootstrap resampling procedure. When comparing the AUC value of the DCNN classifier (0.802) with the pooled AUC value of the three readers (0.729), there was no significant difference (P = 0.079). In differentiating between tuberculous and pyogenic spondylitis using MR images, the performance of the DCNN classifier was comparable to that of three skilled radiologists.