Deep Learning-Assisted Screening and Diagnosis of Scoliosis: Segmentation of Bare Back-Images via an Attention-Enhanced Convolutional Neural Network

Background: Traditional scoliosis screening necessitates a substantial number of specialized personnel and equipment, leading to inconvenience that can result in missed opportunities for early diagnosis and optimal treatment. We have developed a deep learning-based image segmentation model to enhance the efficiency of scoliosis screening. Methods: A total of 350 patients with scoliosis and 108 healthy subjects were included in this study. The dataset comprised bare back images and full-length anteroposterior and lateral X-ray images from 458 participants. An attention mechanism was incorporated into the original U-Net architecture to build an attention U-Net model for image segmentation. The entire dataset was divided into training (321 cases), validation (46 cases), and test (91 cases) sets at a 7:1:2 ratio. The training set was used to train the attention U-Net model, whereas the validation set was used to fine-tune hyperparameters and prevent overfitting during training. The performance of the model was evaluated via the test set. After automatic segmentation of the back contour, a back asymmetry index was calculated viacomputer vision algorithms. The severity of scoliosis was classified on the basis of this index, and the classification results were statistically compared to those of three clinical experts. Results: Following the segmentation of bare-back images and the application of computer vision algorithms, the U-Net model achieved an accuracy, precision, and recall rate of over 90% in predicting severe scoliosis. Notably, the AUC values for diagnosing scoliosis were 0.93 for the U-Net model and 0.92 for the associate chief physician, while for identifying severe scoliosis, the AUC values were 0.95 and 0.96, respectively. Conclusion: The attentionU-Net model developed in this study achieved accuracy and precision in determining scoliosis severity comparable to that of clinical physicians by analyzing bare-back images. The model's ability to diagnose scoliosis was also similar to that ofclinical professionals. The use of this model for scoliosis screening and diagnosis offers advantages such as being radiation-free and improving efficiency. This provides a novel, noninvasive, and effective approach, as well as a theoretical foundation, for large-scale scoliosis screening.