A deep learning model to diagnose and evaluate adolescent idiopathic scoliosis using biplanar radiographs

Background Accurate measurement of the alignment parameters of spinal radiographs is crucial for diagnosing and evaluating adolescent idiopathic scoliosis. Manual measurement is subjective and time-consuming. The recently developed artificial intelligence models mainly focused on measuring the major curve’s CA on the coronal plane and ignored the evaluation of the sagittal plane. Based on that, we developed a deep learning model that could automatically measure alignment parameters in biplanar radiographs. Methods In this study, our model adopted ResNet34 as the backbone network and mainly consisted of landmark detection and CA measurement. A total of 748 biplane radiographs were collected and randomly divided into training and testing sets in a 3:1 ratio. Two senior spinal surgeons independently manually measured alignment parameters and recorded the time taken. The diagnosis performance of the model was evaluated through the ROC curve and AUC. Severity classification and sagittal abnormalities were visualized using a confusion matrix. Compared with the gold standard gold, we tested the reliability and validity of the model using the ICC, simple linear regression, and Bland-Altman plots. Results Our AI model achieved the diagnostic accuracy of scoliosis at 97.2%, and AUC was 0.972 (0.940-1.000). For severity classification, the overall accuracy was 94.5%. All accuracy of sagittal abnormalities was greater than 91.8%. The MAD of coronal and sagittal parameters was 2.15 ° and 2.72 °, and ICC was 0.985, 0.927. The simple linear regression showed a strong correction between all parameters and the gold standard (p < 0.001, R ²  > 0.686), and the Bland-Altman plots showed that the mean difference of the model was within 2 °. Conclusions This deep learning model can accurately and automatically measure spinal alignment parameters with reliable results, significantly reducing diagnostic time, and might provide the potential to assist clinicians.