Scholarlink

<p>This study developed and prospectively validated a deep learning–based automated preoperative planning system for developmental dysplasia of the hip (DDH) to guide osteotomy. Using 3544 anteroposterior pelvic radiographs, a selective keypoint integration strategy combining YOLOv11-Pose, YOLOv11, and YOLOv13 models achieved a mean pixel error of 9.68 pixels with 100% keypoint recall. The system automatically calculates acetabular index, neck-shaft angle, and center–edge angle, and recommends PHP plates based on predefined geometric rules and patient weight. In a multicenter randomized controlled trial with 30 DDH patients and six surgeons, AI planning reduced time from 170.5 ± 4.7&#xa0;s (manual) to 0.0606&#xa0;s (<i>P</i> &lt; 0.001), matched expert acetabular index accuracy (<i>P</i> = 0.327), showed slightly higher but clinically acceptable neck-shaft angle error (<i>P</i> &lt; 0.001), and achieved 92.3% plate recommendation concordance (<i>P</i> = 0.106). This first prospective RCT demonstrates expert-level DDH planning in seconds, advancing AI toward surgical guidance. Clinical trial registration: ChiCTR2600119719 (registered on 2026–03-03).</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

Development and Application of Automated Preoperative Planning System for Developmental Dysplasia of the Hip via Keypoint Selective Integration

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月29日

查看全文
  • Jinghui Yao,
  • Xiaoyou Fan,
  • Yijian Wang,
  • Zijun Gao,
  • Xiuming Huang,
  • Baoxue Sha,
  • Yun Chen,
  • Haiyan Zhang,
  • Daozhang Cai

摘要

This study developed and prospectively validated a deep learning–based automated preoperative planning system for developmental dysplasia of the hip (DDH) to guide osteotomy. Using 3544 anteroposterior pelvic radiographs, a selective keypoint integration strategy combining YOLOv11-Pose, YOLOv11, and YOLOv13 models achieved a mean pixel error of 9.68 pixels with 100% keypoint recall. The system automatically calculates acetabular index, neck-shaft angle, and center–edge angle, and recommends PHP plates based on predefined geometric rules and patient weight. In a multicenter randomized controlled trial with 30 DDH patients and six surgeons, AI planning reduced time from 170.5 ± 4.7 s (manual) to 0.0606 s (P < 0.001), matched expert acetabular index accuracy (P = 0.327), showed slightly higher but clinically acceptable neck-shaft angle error (P < 0.001), and achieved 92.3% plate recommendation concordance (P = 0.106). This first prospective RCT demonstrates expert-level DDH planning in seconds, advancing AI toward surgical guidance. Clinical trial registration: ChiCTR2600119719 (registered on 2026–03-03).