Gong Yuke, Cheng Yan, Liu Yan, Zhang Guohui, Li Shuang, Wu Ruiqi, Wang Hongmei, Lu Lizhou
Ultrasound Medicine Major (Graduate School), Qujing Affiliated Hospital of Kunming Medical University, Qujing, Yunnan, China.
Department of Ultrasound, Dali Prefecture People's Hospital, Dali, Yunnan, China.
J Ultrasound. 2025 Jun 27. doi: 10.1007/s40477-025-01025-8.
This study aims to explore the relationship between ultrasound radiomics features and semantic features from BI-RADS classification in the preoperative differentiation of benign and malignant breast lesions, as well as the potential diagnostic advantages of radiomics features.
Retrospective analysis was performed on 147 female patients with pathologically confirmed breast lesions. Ultrasound images and clinical data were used to construct three diagnostic models: BI-RADS classification single factor diagnostic model, Radiomics diagnostic model, and a BI-RADS-radiomic combined model. Additionally, univariate radiomic models based on semantic features were developed to investigate the associations.
The BI-RADS-Radiomics combined model demonstrated superior performance in both training and testing sets, with AUC values of 0.985 and 0.964, respectively. It also exhibited optimal diagnostic consistency and clinical net benefit. Significant correlations were observed between multiple radiomics features and specific semantic features (AUC range: 0.609-0.752).
Radiomics features effectively assist in breast cancer diagnosis via ultrasound and exhibit nonlinear associations with specific semantic features.
本研究旨在探讨超声影像组学特征与乳腺影像报告和数据系统（BI-RADS）分类中的语义特征在术前鉴别乳腺良恶性病变中的关系，以及影像组学特征的潜在诊断优势。
对147例经病理证实的乳腺病变女性患者进行回顾性分析。利用超声图像和临床数据构建三种诊断模型：BI-RADS分类单因素诊断模型、影像组学诊断模型和BI-RADS-影像组学联合模型。此外，还开发了基于语义特征的单变量影像组学模型来研究其相关性。
BI-RADS-影像组学联合模型在训练集和测试集中均表现出优异的性能，AUC值分别为0.985和0.964。它还表现出最佳的诊断一致性和临床净效益。观察到多个影像组学特征与特定语义特征之间存在显著相关性（AUC范围：0.609-0.752）。
影像组学特征通过超声有效地辅助乳腺癌诊断，并与特定语义特征呈现非线性关联。
