Osteoporotic vertebral compression fractures (OVCFs) are common fragility fractures. Differentiating OVCFs from other vertebra diseases (old fractures (OFs), Schmorl's node (SN), Kummell's disease (KD), previous surgery (PS)) is crucial for treatment. This study developed a deep learning (DL)-driven diagnostic system for multi-type vertebra diseases using computed tomography (CT) images. A two-stage system, comprising a vertebra detection module (VDModule) and a vertebra classification module (VCModule), was developed using a large dataset of CT images (1,051 patients from Luhe Hospital, 46 from Xuanwu Hospital). The VDModule (ResNet18-based Faster R-CNN) achieved an AUC of 0.982, FP rate of 1.52%, and FN rate of 1.33%. The VCModule (ResNet50-based) achieved average sensitivity and specificity of 0.919 and 0.995, respectively, for OVCF, OF, KD, and PS in the testing dataset; 0.891 and 0.989 in the validation dataset. The system accurately diagnoses four vertebra diseases and shows potential for efficient and reliable diagnosis.

Yongjie Wang, Feng, Qian Lu, Wenkai Zhang, Tao Liu, Yining Tao, Shuai Fu, Libin Cui, Shi-Bao B Lu, Xueming Chen, Zhenyun Shi, F Su