实用医学杂志 ›› 2026, Vol. 42 ›› Issue (7): 1192-1200.doi: 10.3969/j.issn.1006-5725.2026.07.011

• 肿瘤诊治与预后专栏 • 上一篇    

基于超声影像组学、O-RADS分类与临床因素的卵巢肿瘤诊断模型

尹晶1,2,张平洋1(),汪珺莉2,尹薇薇2,储小爱2,赵文燕3   

  1. 1.南京医科大学附属南京医院/南京市第一医院心血管超声科 (江苏 南京 210006 )
    2.华东师范大学附属芜湖医院/芜湖市第二人民医院超声医学科 (安徽 芜湖 241000 )
    3.合肥市第一人民院超声医学科 (安徽 合肥 230000 )
  • 收稿日期:2025-12-01 修回日期:2025-12-26 接受日期:2025-12-30 出版日期:2026-04-10 发布日期:2026-04-13
  • 通讯作者: 张平洋 E-mail:zhpy28@126.com
  • 基金资助:
    江苏省卫生健康委科研项目(ZD2021048);皖南医学院校级科研项目(WK2023JXYY133)

Developing an ovarian cancer diagnostic model from ultrasound radiomics, O‑RADS classification, and clinical factors

Jing YIN1,2,Pingyang ZHANG1(),Junli WANG2,Weiwei YIN2,Xiaoai CHU2,Wenyan ZHAO3   

  1. 1.Department of Cardiovascular Ultrasound,Nanjing Hospital Affiliated to Nanjing Medical University,Nanjing First Hospital,Nanjing 210006,Jiangsu,China
    2.Department of Ultrasound Medicine,Wuhu Hospital Affiliated to East China Normal University,Wuhu Second People's Hospital,Wuhu 241000,Anhui,China
    3.Department of Ultrasound Medicine,First People's Hospital of Hefei,Hefei 230000,Anhui,China
  • Received:2025-12-01 Revised:2025-12-26 Accepted:2025-12-30 Online:2026-04-10 Published:2026-04-13
  • Contact: Pingyang ZHANG E-mail:zhpy28@126.com

摘要:

目的 旨在整合超声影像组学、O-RADS(v2022)分类系统及临床危险因素,构建并验证一个智能诊断模型,以提高卵巢肿瘤良恶性鉴别诊断的准确性。 方法 采用多中心回顾性研究设计,纳入华东师范大学附属芜湖医院596例卵巢肿瘤手术患者,按7∶3比例随机分为训练集(n = 418)和内部验证集(n = 178),并收集110例合肥市第一人民院患者作为外部测试集。模型构建包含3个核心组成部分:(1)从标准化超声图像中提取的12个影像组学特征;(2)3位医师盲法评估并达成共识的O-RADS分类结果;(3)经单因素及多因素logistic回归筛选确定的临床预测因子(年龄、肿瘤最大径、CA125、HE4、绝经状态)。研究构建并比较了3种模型:单一O-RADS模型、临床-O-RADS模型以及影像组学-临床-O-RADS综合模型。 结果 联合模型在训练集、内部验证集和外部测试集中的诊断性能均最优,AUC值分别为0.95、0.92和0.89(P < 0.05)。决策曲线分析(DCA)进一步证实,该模型在广泛的阈值概率范围内具有更高的临床净获益。特征重要性分析显示,影像组学特征在模型中贡献度最高(约60%)。 结论 整合超声影像组学、O-RADS分类及临床因素的联合模型能够显著提升卵巢肿瘤的术前鉴别诊断效能,具有良好的泛化能力与临床实用性,为临床决策提供了客观、精准的辅助工具。

关键词: 卵巢肿瘤, 超声诊断, 影像组学, O-RADS分类, 诊断模型

Abstract:

Objective This study aims to integrate ultrasound radiomics, the O-RADS (v2022) classification system, and clinical risk factors to develop and validate an intelligent diagnostic model for improving the accuracy of differentiating between benign and malignant ovarian tumors. Methods A multicenter, retrospective study design was adopted. A total of 596 patients who underwent surgery at our institution were enrolled and randomly split into a training set (n = 418) and an internal validation set (n = 178) at a 7:3 ratio. Additionally, 110 patients from an external hospital were recruited as an external test set. Model construction consisted of three core components: (1) extraction of 12 radiomics features from standardized ultrasound images; (2) O-RADS classification results derived from blinded assessments and consensus among three physicians; (3) clinical predictors (age, maximum tumor diameter, CA125, HE4, and menopausal status) identified through univariate and multivariate logistic regression screening. Three models were developed and compared: a standalone O-RADS model, a combined clinical-O-RADS model, and an integrated radiomics-clinical-O-RADS model. Results The integrated model exhibited the optimal diagnostic performance, with area under the curve (AUC) values of 0.95 in the training set, 0.92 in the internal validation set, and 0.89 in the external test set (P < 0.05). Decision curve analysis (DCA) further confirmed that this model achieved a higher clinical net benefit across a wide range of threshold probabilities. Feature importance analysis revealed that radiomics features contributed the most to the model’s predictive power (approximately 60%). Conclusions The integrated model combining ultrasound radiomics, O-RADS classification, and clinical factors significantly improves the preoperative diagnostic accuracy for distinguishing between benign and malignant ovarian lesions. It demonstrates good generalization ability and clinical utility, providing an objective and precise auxiliary tool to support clinical decision-making in ovarian tumor management.

Key words: ovarian cancer, ultrasound diagnosis, radiomics, O-RADS classification system, diagnostic model

中图分类号: