文献速递关节外科

开发用于全髋关节置换术中假体位置自动化的深度学习模型

Development of a Deep Learning Model for Automating Implant Position in Total Hip Arthroplasty.
PubMed 文献 · J Arthroplasty · 2025-01-27

本研究利用161例初次全髋关节置换术的术前术后骨盆正位X线片和术中透视图像开发深度学习模型,用于标注骨盆标志点并自动测量骨盆倾斜、肢体长度、偏心距、髋臼假体外展角和倾斜角。该模型在20个X线骨性标志点中的16个、4个X线假体标志点中的4个、8个透视骨性标志点中的5个上标注效果不亚于训练有素的人员,并成功计算出杯外展角、前倾角、骨盆倾斜、偏心距和肢体长度。作者认为该模型可快速融入临床实践和科研应用。

要点

  • 深度学习模型利用161例初次全髋关节置换术的影像开发,可标注骨盆标志点并自动测量全髋关节置换相关参数。
  • 模型在X线骨性标志点16/20、X线假体标志点4/4、透视骨性标志点5/8上标注效果不亚于训练有素的人员。
  • 模型成功计算出杯外展角、前倾角、骨盆倾斜、偏心距和肢体长度。

核心结论

  • 作者开发了一种新型深度学习模型,可实时自动识别临床相关的骨盆标志点,并提供与训练有素人员 equivalent 的全髋关节置换相关测量。
  • 作者认为该模型可快速融入临床实践,用于手术和科研应用。

研究设计与样本

  • 研究利用161例初次全髋关节置换术的影像开发深度学习模型,影像包括术前和术后骨盆正位X线片及术中透视图像,并对其进行标注。
  • 设计标志点检测模型对骨盆X线片和透视图像进行标注,算法用于自动测量骨盆倾斜、肢体长度、偏心距、髋臼假体外展角和倾斜角。

主要发现

  • 模型在X线骨性标志点20个中的16个、X线假体标志点4个中的4个、透视骨性标志点8个中的5个上标注效果不亚于训练有素的人员。
  • 模型成功计算出杯外展角和前倾角、骨盆倾斜、偏心距和肢体长度。

临床怎么用

  • 作者认为该模型可快速融入临床实践,用于手术和科研应用。

局限性

  • 原文未报告局限性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

BACKGROUND: Novel methods for annotating antero-posterior pelvis radiographs and fluoroscopic images with deep-learning models have recently been developed. However, their clinical use has been limited. Therefore, the purpose of this study was to develop a deep learning model that could annotate clinically relevant pelvic landmarks on both radiographic and fluoroscopic images and automate total hip arthroplasty (THA)-relevant measurements. METHODS: A deep learning model was developed using imaging from 161 primary THAs. A combination of preoperative and postoperative antero-posterior pelvis radiographs and intraoperative fluoroscopic images were annotated. A landmark detection model was then designed to annotate pelvis radiographs and fluoroscopic images. The algorithm was used to automate the measurement of pelvic tilt, leg length, offset, acetabular component abduction, and inclination. RESULTS: Our novel deep learning model annotated pelvic landmarks as well, if not better, than trained humans at 16 of 20, four of four, and five of eight landmarks for bony landmarks on radiographs, implant landmarks on radiographs, and bony landmarks on fluoroscopy, respectively. Measurements of cup inclination and anteversion, pelvic tilt, offset, and leg length were successfully calculated. CONCLUSIONS: We have developed a novel deep-learning model that can automate the identification of clinically relevant pelvic landmarks in real time and provide THA-relevant measurements that are equivalent to those of trained humans. We believe the model could be rapidly incorporated into clinical practice for both surgical and research applications.

来源:J Arthroplasty,查看原文。版权归原作者所有。

引用

  1. Chan PYW, Baker CE, Suh Y, et al. Development of a Deep Learning Model for Automating Implant Position in Total Hip Arthroplasty. J Arthroplasty. 2025. PMID: 39880053. doi:10.1016/j.arth.2025.01.032

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。