Deep learning-driven diagnosis of multi-type vertebra diseases based on computed tomography images

Background: Osteoporotic vertebral compression fractures (OVCFs) are the most common type of fragility fracture. Distinguishing between OVCFs and other vertebra diseases—old fractures (OFs), Schmorl's node (SN), Kummell's disease (KD), and previous surgery (PS)—is critical for treatment. Leveraging deep learning (DL), this study developed a DL-driven diagnostic system for multi-type vertebra diseases. Methods: A large-scale dataset of sagittal CT images from 1,051 OVCF patients at Luhe Hospital and 46 patients at Xuanwu Hospital (external validation) was constructed, totaling 11,417 CT slices and 19,718 manually annotated diseased vertebrae. A two-stage DL system comprising a vertebra detection module (VDModule) and a vertebra classification module (VCModule) was developed to diagnose five vertebra diseases. Results: The VDModule (ResNet18-based Faster R-CNN) trained on 9,135 and tested on 3,212 vertebrae achieved AUC 0.982, FP 1.52%, and FN 1.33% on the test set. The VCModule (ResNet50 multi-output) trained on 14,584 diseased and 47,604 normal vertebrae and tested on 4,489 diseased and 15,122 normal vertebrae achieved average sensitivity 0.919 and specificity 0.995 for four diseases (OVCF, OF, KD, PS), excluding SN. On the alternative hospital dataset, average sensitivity and specificity were 0.891 and 0.989 for the same four diseases. Conclusions: The proposed DL system accurately diagnoses four vertebra diseases (OVCF, OF, KD, PS) and shows strong potential to facilitate accurate and rapid diagnosis of vertebral diseases.