摘要
基于数字图像处理与机器学习等技术,对直肠癌淋巴结转移情况的诊断问题进行了研究,将肿瘤诊断分解成肿瘤区域提取的图像分割问题与肿瘤区域诊断的图像分类问题.首先,针对肿瘤区域提取的问题,根据直肠肿瘤CT图像的特点,提出了一种结合聚类和水平集方法的图像分割算法,其结果的Dice系数达到0.8954±0.0512,与专业医生人工提取的结果相比具有较高的相似度.然后,使用传统特征提取的方法,针对直肠癌淋巴结是否转移的问题,对肿瘤区域的CT图像进行了分类.实验结果表明,肿瘤的灰度特征与其淋巴结转移情况关联性最高,并且使用PCA降维得到的分类效果最优.最后,本文还使用深度学习方法对肿瘤图像进行了分类.本文使用AlexNet网络模型并采用迁移学习的方法进行训练,实验表明,该方法的效果优于传统方法,其F1-Score达到了0.7719.
The diagnosis of lymph node metastasis of rectal cancer has been studied,which was decomposed into image segmentation problem of tumor region extraction and image classification problem of tumor region diagnosis.First,according to the characteristics of CT images of rectal tumors,an image segmentation algorithm combining clustering and level set method is proposed.The Dice coefficient of the results is 0.8954±0.0512,which is similar to the results of manual extraction by professional doctors.Then,using the traditional feature extraction method,the CT images of the tumor area were classified according to the question of whether the rectal cancer lymph node metastasis.The experimental results show that the gray scale characteristics of the tumor have the highest correlation with lymph node metastasis.The classification effect obtained by PCA dimension reduction is optimal.In addition,the deep learning method is used for the classification of the tumor.We use the AlexNet network model and uses the migration learning method to train.The experiment shows that the method is better than the traditional method.The F1-Score is 0.7719.
作者
吴锐帆
代海洋
杨坦
江颖
蔡志杰
WU Ruifan;DAI Haiyang;YANG Tan;JIANG Ying;CAI Zhijie(School of Data and Computer Science,Sun Yat-sen University,Guangzhou,Guangdong 510275,China;Department of Radiology,Huizhou Municipal Central Hospital,Huizhou,Guangdong 516001,China;School of Mathematical Sciences,South China Normal University,Guangzhou,Guangdong 510631,China;School of Mathematical Sciences,Fudan University,Shanghai 200433,China;Shanghai Key Laboratory for Contemporary Applied Mathematics,Shanghai 200433,China;Key Laboratory of Nonlinear Mathematical Models and Methods of Ministry of Education,Shanghai 200433,China)
出处
《数学建模及其应用》
2019年第4期30-37,共8页
Mathematical Modeling and Its Applications
基金
惠州市科技计划项目(2018Y026).
关键词
直肠癌
淋巴结转移
图像分割
图像分类
深度学习
rectal cancer
lymph node metastasis
image segmentation
feature extraction
AlexNet network