文献速递其他结缔组织病

系统性硬化症手部屈肌腱厚度的自动化测量

Automated Hand Flexor Tendon-Thickness Measurement in Systemic Sclerosis.
PubMed 文献 · Arthritis Care Res (Hoboken) · 2026-08-25

本研究基于多中心HANDSOME队列的超声图像,开发并评估了一种基于nnU-Net的深度学习流程,用于自动分割并测量系统性硬化症(SSc)患者手指屈肌腱厚度。自动化测量在所有图像中成功完成,与专家手动测量具有强一致性(ICC 0.83),平均绝对误差0.25 mm,90%的分割被视觉专家评为临床可接受。

要点

  • 深度学习流程可自动分割屈肌腱、掌骨头和近节指骨基底并测量肌腱厚度。
  • 自动化测量与手动测量一致性强(ICC 0.83),平均绝对误差0.25 mm。
  • 90%的自动分割被视觉专家评为临床可接受。

核心结论

  • 自动化分析纵向手部超声图像能够准确测量屈肌腱厚度,可能促进标准化肌腱评估并支持肌腱受累研究。

研究设计与样本

  • 概念验证研究,超声图像来自多中心HANDSOME队列,训练集包括59名个体的687张图像,独立测试集包括15名个体的150张图像。
  • 开发了基于nnU-Net的深度学习流程,自动分割屈肌腱、掌骨头和近节指骨基底,定位掌指关节后垂直于肌腱自动测量厚度。
  • 自动化测量与专家手动测量进行比较。

主要发现

  • 自动化肌腱厚度测量在所有图像中成功获得(100%)。
  • 模型分割性能高,屈肌腱Dice相似系数为0.90,近节指骨基底为0.83,掌骨头为0.76。
  • 自动化厚度测量与手动测量一致性强,组内相关系数为0.83。
  • 平均绝对测量误差为0.25 mm,相对测量误差为8.0%。
  • 视觉专家评估将90%的自动分割分类为临床可接受。

临床怎么用

  • 该自动化方法可用于SSc患者手部超声图像的标准化屈肌腱厚度评估,减少手动测量的时间和观察者间变异。

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

OBJECTIVE: Systemic sclerosis (SSc) can affect flexor tendons, contributing to hand function problems and reduced quality of life. Tendon changes are currently assessed with ultrasonography and measured manually, a time-consuming process prone to interobserver variability. This study aimed to develop and evaluate an automated method for standardized measurement of finger flexor tendon thickness on longitudinal ultrasonography images in patients with SSc. METHODS: In this proof-of-concept study, ultrasonography images were obtained from the multicenter HANDSOME cohort. The training dataset included 687 images from 59 individuals. The independent testing dataset consisted of 150 images from 15 individuals. A deep learning-based pipeline using nnU-Net was developed to automatically segment the flexor tendon, metacarpal head, and proximal phalanx base. The metacarpophalangeal joint was subsequently localized, after which tendon thickness was automatically measured perpendicular to the tendon. Automated measurements were compared with manual expert measurements. RESULTS: Automated tendon-thickness measurements were successfully obtained in all images (100%). The model demonstrated high segmentation performance, with Dice similarity coefficients of 0.90 for the flexor tendon, 0.83 for the proximal phalanx base, and 0.76 for the metacarpal head. Automated thickness measurements showed strong agreement with manual measurements (intraclass correlation coefficient 0.83). The mean absolute measurement error was 0.25 mm, corresponding to a relative measurement error of 8.0%. Visual expert evaluation classified 90% of automated segmentations as clinically acceptable. CONCLUSION: Automated analysis of longitudinal hand ultrasonography images enables accurate measurement of flexor tendon thickness. This approach may facilitate standardized tendon assessment and support studies investigating tendon involvement.

来源:Arthritis Care Res (Hoboken),查看原文。版权归原作者所有。

引用

  1. Greveling M, van Het Erve T, van de Berg Y, et al. Automated Hand Flexor Tendon-Thickness Measurement in Systemic Sclerosis. Arthritis Care Res (Hoboken). 2026. PMID: 42640177. doi:10.1002/acr.80149

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。