Scholarlink

Background <p>Adolescent idiopathic scoliosis (AIS) affects 2–3% of adolescents. Current screening relies on X-rays, which limits large-scale application due to radiation exposure and equipment requirements. This study aimed to develop a radiation-free, efficient screening method using bare-back images and an efficient dilated residual network (EDRNet).</p> Methods <p>In this study, 300 adolescent patients with scoliosis and 300 adolescent patients without spinal diseases were selected from among those treated at the Department of Orthopedics of Ningxia Medical University General Hospital and the Department of Spinal Surgery of Henan Provincial People’s Hospital between January 2021 and December 2024. Their bare-back images were used as the study subjects. The dataset was expanded using data augmentation techniques, and an efficient dilated residual network was proposed. Five classification models for scoliosis based on back images were constructed: VGG16, GoogLeNet, ResNet34, MobileNetV2, and the EDRNet proposed in this study. To evaluate the performance of the models, accuracy, precision, recall, and F1 score were used to analyze and compare the prediction results. Additionally, ROC curves and bubble charts were used to compare the classification ability of EDRNet with that of clinicians on bare-back images.</p> Results <p>When using both original and augmented data for predictive analysis, all five deep learning models demonstrated good predictive performance. Among them, the proposed EDRNet model exhibited the best predictive performance across all models. In comparing the classification ability on bare-back images, EDRNet (AUC = 0.91) was lower than that of the deputy chief physician (AUC = 0.93) but higher than that of the attending physician (AUC = 0.88) and the resident physician (AUC = 0.86), and EDRNet’s speed was superior to that of the three physicians.</p> Conclusions <p>The scoliosis classification model based on EDRNet shows good predictive accuracy and can achieve efficient scoliosis screening through the analysis of bare-back images. It has broad application prospects, especially in large-scale adolescent scoliosis screening.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

EDRNet-based adolescent idiopathic scoliosis screening method using bare-back images

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月16日

查看全文
  • Xingyu Duan,
  • Linan Wang,
  • Zhengyong Tao,
  • Mengqi Zhu,
  • Dingqi You,
  • Yuxin Gao,
  • Zongqiang Yang,
  • Wensheng Liao,
  • Ningkui Niu

摘要

Background

Adolescent idiopathic scoliosis (AIS) affects 2–3% of adolescents. Current screening relies on X-rays, which limits large-scale application due to radiation exposure and equipment requirements. This study aimed to develop a radiation-free, efficient screening method using bare-back images and an efficient dilated residual network (EDRNet).

Methods

In this study, 300 adolescent patients with scoliosis and 300 adolescent patients without spinal diseases were selected from among those treated at the Department of Orthopedics of Ningxia Medical University General Hospital and the Department of Spinal Surgery of Henan Provincial People’s Hospital between January 2021 and December 2024. Their bare-back images were used as the study subjects. The dataset was expanded using data augmentation techniques, and an efficient dilated residual network was proposed. Five classification models for scoliosis based on back images were constructed: VGG16, GoogLeNet, ResNet34, MobileNetV2, and the EDRNet proposed in this study. To evaluate the performance of the models, accuracy, precision, recall, and F1 score were used to analyze and compare the prediction results. Additionally, ROC curves and bubble charts were used to compare the classification ability of EDRNet with that of clinicians on bare-back images.

Results

When using both original and augmented data for predictive analysis, all five deep learning models demonstrated good predictive performance. Among them, the proposed EDRNet model exhibited the best predictive performance across all models. In comparing the classification ability on bare-back images, EDRNet (AUC = 0.91) was lower than that of the deputy chief physician (AUC = 0.93) but higher than that of the attending physician (AUC = 0.88) and the resident physician (AUC = 0.86), and EDRNet’s speed was superior to that of the three physicians.

Conclusions

The scoliosis classification model based on EDRNet shows good predictive accuracy and can achieve efficient scoliosis screening through the analysis of bare-back images. It has broad application prospects, especially in large-scale adolescent scoliosis screening.