Abstract: Objective To explore the value of color doppler flow imaging（CDFI） imaging radiomics signature in the diagnosis of benign and malignant breast nodules. Methods The data of 159 patients （161 cases in total） with breast nodules detected through breast color doppler ultrasound were retrospectively collected. All cases were confirmed by pathology. The 161 cases were divided into group A and group B （98 cases in group A and 63 cases in group B） according to the sample size of benign and malignant. According to the sample size of 7∶3, benign （69∶29） and malignant （44∶19） nodes were randomly selected as the training set （113 cases） and the validation set （48 cases）. For all cases, 1125 features were extracted for each lesion based on the CDFI images of the nodules using the Darwin Research Platform for lesion outlining and feature extraction. Then, the F-classif of the sample variance was used to downscale and filter the features, from which 10 optimal features were selected, and the CDFI-based radiomics signature was established by the logistic regression mode. Usingreceiver′s operating characteristic （ROC） curve to evaluate the efficiency of differential diagnosis between benign and malignant breast nodules. Results The AUC of ultrasound radiomics signature was 0.895（95%CI：0.838-0.951） in training set and 0.904（95%CI：0.801-1.000） in testing set, and theradiomics signature was related to the differential diagnosis of benign and malignant breast nodules （P<0.001）. Conclusion The color doppler ultrasound-based radiomics signature has high efficiency in differential diagnosis of benign and malignant breast nodules, and it can assist the clinic in providing new imaging ideas for early diagnosis of breast nodules, the formulation of treatment plans, and the non-invasive assessment of breast cancer prognosis before surgery.

Key words: breast nodules, radiomics signature, color doppler ultrasound imaging, differential diagnosis

摘要： 目的 探讨基于彩色多普勒超声图像的影像组学标签在乳腺结节良恶性鉴别诊断中的价值。方法 回顾性分析通过乳腺彩色多普勒超声（CDFI）检查,发现乳腺结节患者159人（共161个病灶）,所有病例均经病理确诊。这161个病灶按照样本量的良性和恶性分为A组和B组（A组有98例患者,B组有63例）,每组按7∶3随机抽样选取良性结节（69∶29）及恶性结节（44∶19）作为训练集（共113例）和验证集（共48例）。对于所有病例,基于结节的CDFI图像,使用达尔文科研平台对病灶进行勾画和特征提取,每个病灶提取了1125个特征。然后,采用样本方差F值对特征进行降维和筛选,从中选出了10个最优特征,并利用逻辑回归模型构建组学标签鉴别结节良恶性。使用受试者工作特征（ROC）曲线对乳腺良恶性结节进行鉴别诊断效能评估。结果 超声影像组学标签在训练集中的AUC为0.895（95%CI：0.838~0.951）,测试集的AUC为0.904（95%CI：0.801~1.000）,标签与乳腺良恶性结节的鉴别诊断相关（P<0.001）。结论 基于彩色多普勒超声构建的影像组学标签对乳腺良恶性结节具有较高的鉴别诊断效能,可辅助临床为病灶的早期诊断、治疗方案的制定及术前无创评估乳腺癌预后提供影像学新思路。

关键词: 乳腺结节, 影像组学特征, 彩色多普勒超声图像, 鉴别诊断