Answers are grounded in this paper and cite their sources.
Background Cobb angle and vertebral slip measurements are clinically vali-dated techniques that are time-intensive and prone to inter-observer variation. Objective To develop a comprehensive deep learning approach for auto-mated measurements and grading of scoliosis and spondylolisthesis cases using a single-center hospital radiograph dataset. Methods In this retrospective study, 2,500 anonymized spinal radiographs from one participating hospital were used, including 1,635 scoliosis and 865 spondy-lolisthesis radiographs. Researchers had access only to radiographs without any patient identifiers, demographic information, and clinical data. Radiographs were preprocessed by applying Gaussian denoising, CLAHE contrast adjust-ment, intensity standardization, and scaling to 224 x 224 pixels. Vertebral landmark-based vertebral angles were computed for scoliosis analysis, and verte-bral body localization-based slip percentages were calculated for spondylolisthesis detection. Two networks, DenseNet201, and ResNet152, were fine-tuned with two-stage ImageNet transfer learning approach. Results DenseNet201 obtained an accuracy of 87.5% for scoliosis and 88.4% for spondylolisthesis on hold-out test subsets. Cobb-angle MAE was 1.845° and slip percentage MAE was 1.8 percentage point. The corresponding ICC values were 0.962 and 0.958, and macro-AUC values were 0.83 and 0.84. Conclusion The proposed method demonstrated strong internal validity in terms of spinal deformity measurement and grading within the context of a single-center study. Since patient matching was not possible, patient-level independence of the image split could not be confirmed. These results, therefore, constitute internal validation at the level of the image and need to be externally validated independently. Clinical trial registration Not applicable. This study was retrospective and did not involve a clinical trial or prospective intervention.
No takes yet. Share an insight, caveat, or question.
Bhawana et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: