摘要
目的基于人工智能技术的前列腺癌诊断是目前的研究热点。然而现有的智能诊断方法大多只能利用MRI、CT等图片数据进行前列腺癌诊断,还无法对前列腺癌的图数据进行处理,局限性很大,诊断性能还有待提高。为了弥补这一不足,提出一种基于图卷积神经网络(GCN)的前列腺癌诊断模型(PCa-GCN)。方法首先从各大医院收集到前列腺癌检查数据样本,然后基于jieba分词、词袋模型、最大熵模型等预处理来构建得到病历图。接着将病历图作为GCN的输入来学习得到前列腺癌特征的图嵌入表示,最后采用基于Sigmoid的逻辑回归实现特征与前列腺癌之间的映射,完成对前列腺癌的准确诊断。结果基于K折交叉验证的实验结果表明,PCa-GCN模型在召回率和ROC曲线方面的性能优于其他诊断方法。结论PCa-GCN模型可实现前列腺癌的精准诊断,为前列腺癌数据分析、疾病预防提供技术支持。
Objective Diagnosis of prostate cancer based on artificial intelligence technology attracts great academic attention at present.However,most of the existing intelligent diagnostic methods can only collect MRI,CT and other image data for prostate cancer diagnosis,and cannot process such data,resulting in huge limitations and unsatisfactory performance.In order to solve the problem,a prostate cancer diagnosis model based on graph convolutional neural network(PCa-GCN)is proposed in this paper.Methods Firstly,prostate cancer data samples were collected from various hospitals,and then the graph of medical record was constructed based on the preprocessing with the jieba word segmentation,the bag-of-words model and the maximum entropy model.Then the graph of medical record was inputted for GCN to learn the graph embedding representation of characteristics of prostate cancer.Finally,the mapping between such features and prostate cancer was accomplished by logic regression based on sigmoid for accurate diagnosis of prostate cancer.Results Experimental results based on k-fold cross validation showed that the PCa-GCN model is superior to the other diagnostic methods in terms of recall rate and ROC curve.Conclusion PCa-GCN model achieved precise diagnosis of prostate cancer and can provide technical support for prostate cancer data analysis and disease prevention.
作者
李鹏
罗爱静
闵慧
Li Peng;Luo Aijing;Min Hui(The Third Xiangya Hospital of Central South University,Changsha 410013,China;School of Informatics,Hunan University of Chinese Medicine,Changsha 410208,China;CSU Key Laboratory of Medical Information Research of Colleges and Universities in Hunan Province,Changsha 410006,China;School of Software,Hunan College of Information,Changsha 410200,China)
出处
《北京中医药大学学报》
CAS
CSCD
北大核心
2020年第12期1034-1041,共8页
Journal of Beijing University of Traditional Chinese Medicine
基金
国家重点研发计划项目(No.2017YFC1703306)
湖南自然科学基金青年项目(No.2019JJ50453)
湖南自然科学基金面上项目(No.2018JJ2301)
湖南省科技厅重点项目(No.2017SK2111)
湖南中医药大学开放基金项目(No.2018JK02)
湖南省教育厅一般项目(No.19C1318)。
关键词
前列腺癌
图卷积神经网络
病历图
图嵌入
诊断
召回率
prostate cancer
graph convolutional network
graph of medical record
graph embedding representation
diagnosis
recall rate
作者简介
李鹏.男,博士,讲师;通信作者:罗爱静,女,教授,博士生导师,主要研究方向:医药信息管理、卫生信息符理、医药信息检索,E-mail:lpch617@csu.edu.cn。
