基于双序列MRI的计算机辅助诊断鉴别软组织肿瘤良恶性的研究  

MRI-based computer-aided differentiation between benign and malignant soft-tissue tumors

作　　者：姜文研 尚圣捷 王颖妮 王晓煜[1] 宋江典 龙哲[2] 于韬[1] 罗娅红[1] Jiang Wenyan;Shang Shengjie;Wang Yingni;Wang Xiaoyu;Song Jiangdian;Long Zhe;Yu Tao;Luo Yahong(Cancer Hospital of China Medical University,Liaoning Cancer Hospital&Institute Liaoning,Liaoning Shenyang110042,China;Department of Biomedical Engineering,China Medical University,Liaoning Shenyang 110122,China;College of Medical Informatics,China Medical University,Liaoning Shenyang 110122,China)

机构地区：[1]辽宁省肿瘤医院(中国医科大学肿瘤医院),辽宁沈阳110042 [2]中国医科大学生物医学工程系,辽宁沈阳110122 [3]中国医科大学医学信息学院,辽宁沈阳110122

出　　处：《现代肿瘤医学》2020年第12期2122-2126,共5页Journal of Modern Oncology

摘　　要：目的:研究使用计算机辅助分析方法针对软组织肿瘤MRI影像进行肿瘤良恶性鉴别的价值。方法:回顾性收集了在辽宁省肿瘤医院就诊的72例软组织肿瘤患者的CE-T1和T1WI双序列MRI影像数据(2017年1月至2018年1月)。通过提取和筛选MRI影像特征,建立支持向量机(SVM)、K-最邻近(KNN)和随机森林(RF)三种机器学习分类器模型对肿瘤病灶进行二分类鉴别;提出一种新型集成学习分类器模型用于将两个序列MRI信息进行融合。通过绘制受试者工作特征曲线(ROC)并计算ROC曲线下面积(AUC值)以评估模型的分类鉴别能力。结果:三种机器学习分类器均取得较好的良恶性鉴别效果;提出的集成学习分类器的分类效果最佳,AUC值达到0.922(敏感性=0.965,特异性=0.783)。结论:本研究提出的计算机辅助模型能够利用MRI影像对软组织肿瘤良恶性进行有效的辅助鉴别,具有一定的潜在应用价值。Objective:To analyze the identification of benign and malignant tumors using MRI images of soft tissue tumors by computer-aided diagnosis.Methods:CE-T1 and T1 WI dual-sequence MRI image data were retrospectively collected from 72 soft-tissue tumors patients from Liaoning Provincial Cancer Hospital from Jan.2017 to Jan.2018.After extracting and selecting MRI image features,support vector machine(SVM),K-nearest neighbor(KNN)and random forest(RF)classifiers were established to discriminate tumor lesions.An ensemble learning classifier was proposed integrating both the CE-T1 and T1 WI MRI.Operating characteristic curves(ROC)were plotted to assess the predictive abilities of the models with the area under the ROC curve(AUC).Results:The three machine learning classifiers all generated acceptable prediction results.The proposed ensemble learning classifier gave a highest AUC value of 0.922(sensitivity=0.965,specificity=0.783).Conclusion:The proposed ensemble learning classifier based on multimodal MRI data of soft-tissue tumors in this study has great potential in non-invasive distinguishing malignant from benign tumors.

关 键 词：软组织肿瘤 MRI 计算机辅助分析 

分 类 号：R738.6[医药卫生—肿瘤]