文献速递其他结缔组织病

腮腺超声影像组学检测干燥综合征

Detecting Sjögren's Disease from Parotid Gland Ultrasound Radiomics.
PubMed 文献 · Rheumatology (Oxford) · 2026-08-11

本研究纳入202名参与者的866张腮腺超声图像,提取104个影像组学特征并训练五折软投票SVM集成模型,区分干燥综合征(SjD)患者与健康对照的AUC达0.99,准确率0.94,优于放射科医生视觉评估的0.62和0.72,但非干燥综合征口干患者与确诊SjD患者的影像组学特征存在明显重叠。

要点

  • 腮腺超声影像组学区分SjD与健康对照AUC达0.99,准确率0.94。
  • 模型性能优于放射科医生视觉评估(准确率0.62和0.72)。
  • 非干燥综合征口干患者与确诊SjD患者的影像组学特征存在明显重叠。

核心结论

  • 基于腮腺超声的影像组学机器学习在区分SjD与健康对照方面表现出高分类性能。
  • 定量超声分析可能作为SjD评估的客观辅助工具,但需在更大规模多中心队列中验证。

研究设计与样本

  • 研究纳入202名参与者的866张腮腺超声图像,包括123例符合2016 ACR/EULAR标准的SjD患者、33例健康对照、24例非干燥综合征口干患者和22例不完全SjD病例。
  • 提取104个描述强度、纹理和微纹理模式的影像组学特征,在确诊SjD与健康参与者上训练五折软投票SVM集成模型。
  • 非干燥综合征口干和不完全SjD病例保留为独立测试集,采用SHAP分析进行模型可解释性分析。

主要发现

  • SVM集成模型区分SjD与健康对照的AUC为0.99,准确率0.94,敏感性0.86,特异性0.96。
  • 模型性能优于放射科医生视觉评估(准确率分别为0.62和0.72)。
  • SHAP分析识别强度离散度指标、基于GLCM的纹理特征和LBP微纹理模式为最强预测因子。
  • PCA和特征层面分析显示非干燥综合征口干患者与确诊SjD患者的影像组学特征存在显著重叠。

关键数值与推荐

  • SjD诊断采用2016 ACR/EULAR标准。
  • SVM集成模型区分SjD与健康对照的AUC为0.99,准确率0.94,敏感性0.86,特异性0.96。
  • 放射科医生视觉评估准确率分别为0.62和0.72。

临床怎么用

  • 定量腮腺超声影像组学分析可能作为SjD评估的客观辅助工具,帮助减少操作者依赖性。
  • 由于非干燥综合征口干患者与确诊SjD患者影像组学特征存在重叠,该工具在鉴别口干病因时需谨慎解读。

局限性

  • 需要在更大规模多中心队列中验证。
  • 非干燥综合征口干患者与确诊SjD患者的影像组学特征存在显著重叠。

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

OBJECTIVES: Salivary gland ultrasonography is a promising non-invasive modality for the evaluation of Sjögren's disease (SjD), but its diagnostic utility is limited by operator dependency. This study aimed to evaluate the classification performance of radiomics-based machine learning using parotid gland ultrasonography and to compare it with conventional visual assessment. METHODS: A total of 866 parotid gland ultrasound images from 202 participants were included: 123 patients fulfilling the 2016 ACR/EULAR criteria for SjD, 33 healthy controls, 24 non-Sjögren sicca patients, and 22 incomplete SjD cases. A total of 104 radiomic features describing intensity, texture, and micro-texture patterns were extracted. A 5-fold soft-voting SVM ensemble was trained on confirmed SjD and healthy participants; non-Sjögren sicca and incomplete SjD cases were reserved for the held-out test set. SHAP analysis was used for model interpretability. RESULTS: The SVM ensemble achieved an area under the receiver operating characteristic curve (AUC) of 0.99 for binary classification between SjD and healthy controls, with 0.94 accuracy, 0.86 sensitivity, and 0.96 specificity, outperforming radiologist assessments (accuracy: 0.62 and 0.72). SHAP analysis identified intensity dispersion metrics, GLCM-based texture features, and LBP micro-texture patterns as the strongest predictors. PCA and feature-level analyses demonstrated substantial overlap in radiomic features between non-Sjögren sicca and confirmed SjD patients. CONCLUSION: Radiomics-based machine learning demonstrated high classification performance for distinguishing SjD from healthy controls using parotid gland ultrasonography. Quantitative ultrasound analysis may serve as an objective adjunctive tool in SjD assessment, although validation in larger multicenter cohorts is required.

来源:Rheumatology (Oxford),查看原文。版权归原作者所有。

引用

  1. Akkuzu G, Durugol OF, Tolu S, et al. Detecting Sjögren's Disease from Parotid Gland Ultrasound Radiomics. Rheumatology (Oxford). 2026. PMID: 42581604. doi:10.1093/rheumatology/keag409

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。