基于深度卷积神经网络的数字病理辅助诊断系统设计  

Design of the Digital Pathology-assisted Diagnosis System Based on the Deep Convolutional Neural Network

作　　者：王继伟 樊伟 陈岗 陈福沨 方斌 雷晓晔 杨文圣 WANG Ji-wei;FAN Wei;CHEN Gang(The 73rd Group Army Hospital(Chenggong Hospital Affiliated to Xiamen University),Xiamen 361003,Fujian Province,P.R.C.)

机构地区：[1]陆军第七十三集团军医院(厦门大学附属成功医院),福建省厦门市361003 [2]福州数据技术研究院,福建省福州市350200

出　　处：《中国数字医学》2020年第12期48-52,共5页China Digital Medicine

基　　金：厦门市科技计划(编号:3502Z20209154)。

摘　　要：目的:帮助病理医生快速定位病变区域,提高诊断效率,减少漏诊。方法:提出基于深度卷积神经网络的病理辅助诊断方法,通过图像预处理对病理图像提取前景和染色归一化,采用病理医生标注的数据集训练和优化模型的结构和参数,最后使用训练后的神经网络对病理图像进行正常组织和疑似病变组织二分类识别。结果:以乳腺癌组织病理影像为例对该方法进行测试,识别结果达到了95%以上的灵敏度和特异度,系统的诊断准确率为95.8%。结论:数字化智能病理诊断系统集成的智能算法模块,达到较高的特异度和灵敏度,使人工智能辅助诊断能更好地介入到病理诊断工作中去,实现了数字化病理智能辅助诊断,为病理医生提供实际的业务帮助。Objective:To help pathologists with early detection of diseases,improve diagnosis efficiency and reduce the risk of misdiagnosis.Methods:A computer-aided pathological diagnosis system based on DCNN was designed.Key modules involve image preprocessing,including foreground detection and color normalization of pathological images,and model training,where DCNN is trained and optimized by pathological images labeled by pathologists.Finally,normal and suspected abnormal tissues are detected and classified by the trained artificial neural network.Results:When this system is tested by histopathological images of breast cancer,both specificity and sensitivity are over 95%,and diagnostic accuracy is 95.8%.Conclusion:Intelligent algorithm module in this system has achieved high specificity and sensitivity,indicating it has the potential to be integrated to digital pathological diagnosis in practice and provide operational assistance for pathologists.

关 键 词：数字病理 人工智能 辅助诊断系统 

分 类 号：R319[医药卫生—基础医学] TP391[自动化与计算机技术—计算机应用技术]