A new deep learning pipeline was developed to help identify spinal fractures and Schmorl's nodes in older adults. The system was tested on 5,000 chest or abdominal CT scans acquired for non-spinal indications in routine clinical care. It compared its findings against radiologist consensus to see how well it could segment vertebrae and detect specific injuries.
The pipeline showed strong performance across multiple tasks. It achieved 96% accuracy for anatomical labeling when at least one spinal extremity was visible. When both extremities were absent, accuracy dropped to 86%. For top-three anatomical labeling, the system reached 100% accuracy in both settings. It also correctly identified fractured vertebrae in 92.7% of cases with 92.6% sensitivity and 95.8% specificity.
The system also graded fracture severity with 85% overall accuracy and detected Schmorl's nodes with 91.9% accuracy. It matched radiologist consensus 90% of the time. This work provides a practical framework for integrating AI-assisted vertebral screening into routine radiology practice. No safety concerns were reported because this was a development study without patient treatment.
readers should understand this is a technical framework for AI integration rather than a clinical trial. The results show high accuracy but do not prove the tool is ready for immediate use without further validation. This guideline supports the idea of using such tools to assist radiologists in spotting spinal issues during standard scans.