摘要: |
目的 基于灰阶超声(US)与剪切波弹性(SWE)的双模态图像与临床病理资料构建乳腺癌腋窝淋巴结转移预测模型。方法 回顾性分析306名2018年6月至2021年12月于南京鼓楼医院接受乳腺癌手术治疗的患者,并按照7:3比例随机划分为训练队列(n = 214)和验证队列(n = 92)。基于术前常规US和SWE,分别进行感兴趣区域(ROI)分割与特征提取。运用最小绝对收缩和选择算法(LASSO)筛选关键特征并分别构建US与SWE影像组学标记物(RIS)。采用单因素和多因素Logistics回归在临床病理资料和RIS中筛选变量并构建模型。采用受试者操作特征曲线(ROC)的曲线下面积(AUC)评价并比较双模态模型与单模态模型和超声医师的预测或判断效能。 结果 基于LASSO筛选的13个US特征与17个SWE特征,构建US-RIS与SWE-RIS。联合BI-RADS分级、肿瘤分类、US-RIS与SWE-RIS构建的双模态Logistics预测模型,在训练队列和验证队列中,其AUC、准确率、灵敏度、特异度、分别为0.926、84%、78%、88%和0.863、78%、62%、88%,均高于单模态US模型、单模态SWE模型与超声医生的腋窝淋巴结评估结果(P<0.05)。结论 联合US、SWE图像与临床病理信息建立的双模态影像组学模型可以确定乳腺癌腋窝淋巴结转移风险,在术前非侵入性地为个体化治疗提供参考并避免过度的前哨淋巴结活检。 |
关键词: 乳腺癌 剪切波弹性成像 超声 影像组学 列线图 |
DOI: |
投稿时间:2022-08-25修订日期:2023-02-24 |
基金项目:南京市医学科技发展资金(QRX17011) |
|
Construction of a Bimodal Radiomics Model for Axillary Lymph Node Metastasis of Breast cancer based on Gray Scale Ultrasound and Shear Wave Elastography |
liuhan,xunan,wujie,zhangyidan,wangying,shenhaiyun,kongwentao |
(Department of Ultrasound, The Affiliated Drum Tower Hospital of Nanjing University Medical School) |
Abstract: |
Objective To develop a bimodal radiomics model integrating gray scale ultrasound (US), shear wave elastography (SWE), and clinicopathological characteristics for axillary lymph node (ALN) metastasis prediction in breast cancer. Methods A retrospective analysis was performed on 306 patients who underwent breast cancer surgery in Nanjing Drum Tower Hospital from June 2018 to December 2021.They were divided into a training cohort (n = 214) and a test cohort (n = 92) randomly in a ratio of 7:3. Region of interest (ROI) segmentation and feature extraction were performed based on preoperative US and SWE, respectively. The least absolute shrinkage and selection operator (LASSO) algorithm was used to screen key features and construct radiomic biomarkers (RIS) of US and SWE, respectively. Univariate and multivariate logistics regression were utilized to construct the radiomics model based on clinicopathological data and RIS. The area under the receiver operating characteristic curve (AUC) was used to evaluate the incremental value of the bimodal radiomic model than the two single-mode models and the sonographer. Results Based on 13 US features and 17 SWE features screened by LASSO, the RIS of US and SWE were constructed. Combined with BI-RADS grading, tumor classification, US-RIS and SWE-RIS, the dual-mode logistics prediction model was constructed. For the training and test set, the AUC, accuracy, sensitivity and specificity of the model were 0.926, 84%, 78%, 88%, and 0.863, 78%, 62%, 88%, respectively. The performance of the dual-mode model was better than those of the single mode US model, the single mode SWE model and the sonographers(P<0.05). Conclusions A dual-mode radiomics model combined with US and SWE images and clinicopathological information can identify the risk of axillary lymph node metastasis of breast cancer. It can provide a non-invasive method to guide individualized treatment preoperatively, and to avoid excessive sentinel lymph node biopsy. |
Key words: Deep learning, Radiomics, Artificial intelligence, Breast cancer, Ultrasound |