基于超声影像组学及临床特征构建乳腺癌列线图的预测模型
A prediction model of breast cancer nomogram based on ultrasound radiomics and clinical features
张盼盼 1孙医学 1李阳 1李林 1杜欢 1路丽丽 1乔佳业1
作者信息
- 1. 蚌埠医学院第一附属医院超声医学科,安徽 蚌埠 233017
- 折叠
摘要
目的:通过整合临床危险因素和术前超声影像组学评分,建立一个基于超声影像组学的列线图预测乳腺癌.方法:回顾性收集2020年10月—2023年2月有明确病理结果的525例患者的525个乳腺肿块(其中良性241例,恶性284例)的超声图像,按照7∶3比例随机分为训练组(368例)、验证组(157例).根据肿块轮廓勾画出肿瘤的感兴趣区域(Region of interest,ROI),并提取影像组学特征.采用最小绝对收缩和选择算子(Least absolute shrinkage and selection operator,LASSO)对超声影像组学特征进行降维分析,选择Logistic回归分类器将结果输出转换为影像组学评分(Radiomics score,Rad-Score),作为Rad-Score模型.此外,采用Logistic回归方法将影像组学评分与临床危险因素进行整合,构建联合模型并绘制列线图.绘制ROC曲线及校准曲线以评价模型效能.结果:提取的851个影像组学特征中筛选出13个非零特征用于建立模型,多因素分析中,乳腺癌患者的独立危险因素是年龄,基于患者年龄及Rad-Score构建联合模型,在训练集和验证集中,临床模型的AUC值分别为:0.772、0.847;影像组学模型的AUC值分别为:0.790、0.820;联合模型的AUC值分别为:0.846、0.909.DeLong检验显示在训练集和验证集中,联合模型优于临床模型(P＜0.05),联合模型与影像组学模型及影像组学模型与临床模型之间差异无统计学意义(P＞0.05).训练集和验证集中联合模型的校准曲线可以看出其预测乳腺癌风险概率与实际发生率较为接近,可以更好地指导临床决策.结论:基于临床危险因素及超声影像组学模型构建的列线图,对于预测乳腺癌有较高的价值.
Abstract
Objective:To integrating clinical risk factors and preoperative echotomic scores,to develop an echotomic-based nomogram to predict breast cancer.Methods:Ultrasound images of 525 breast masses(241 benign and 284 malignant)from 525 patients with definite pathological results from October 2020 to February 2023 were retrospectively collected.They were randomly divided into training group(368 cases)and verification group(157 cases)according to the ratio of 7∶3.The region of interest(ROI)was delineated according to the outline of the tumor,and the radiomics features were obtained.After dimension reduction analysis using the least absolute shrinkage and selection operator(LASSO),and the logistic regression classifier was selected to convert the resulting output to the radiomics score(Rad-Score)as the Rad-Score model.In addition,logistic regression was used to integrate radiomics scores with clinical risk factors to construct a combined diagnostic model and draw a nomogram.ROC curves and calibration curves were drawn to evaluate the model performance.Results:Of the 851 radiomics features extracted,13 non-zero features were selected for building the Rad-Score model.In multivariate analysis,age was the independent risk factor for breast cancer patients.A combined model was constructed based on age and Rad-Score.In the training group and validation group,the AUC values of clinical model were 0.772 and 0.847;the AUC values of radiomics model were 0.790 and 0.820;the AUC values of combined model were 0.846,0.909.The DeLong test showed that in the training set and validation set,the combined model was better than the clinical model(P＜0.05),and there was no significant difference between the combined model and the radiomics model,as well as between the radiomics model and the clinical model(P＞0.05).The calibration curve of the combined model in the training set and the validation set showed that the probability of predicting breast cancer risk was closed to the actual incidence,which could better guide clinical decision-making.Conclusion:The nomogram constructed based on clinical risk factors and ultrasound radiomics score has high value for predicting breast cancer.
关键词
乳腺肿瘤/超声检查,乳房/危险因素/人工智能/影像组学/列线图Key words
Breast cancer/Ultrasonography,Breast/Risk factors/Artificial intelligence/Radiomics/Nomogram引用本文复制引用
基金项目
蚌埠医学院自然科学重点科技项目(2021byzd066)
出版年
2024
国家科技期刊平台
NSTL
万方数据