基于改进深度卷积神经网络的病理图像有丝分裂检测算法研究  

MITOSIS DETECTION ALGORITHM IN PATHOLOGICAL IMAGE BASED ON IMPROVE DEEP CONVOLUTION NEURAL NETWORK

作　　者：齐莹 刘振丙[1] 潘细朋 杨辉华[1,2] Qi Ying;Liu Zhenbing;Pan Xipeng;Yang Huihua(School of Electronic Engineering and Automation,Guilin University of Electronic Technology,Guilin 541004,Guangxi,China;School of Automation,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]桂林电子科技大学电子工程与自动化学院,广西桂林541004 [2]北京邮电大学自动化学院,北京100876

出　　处：《计算机应用与软件》2018年第9期199-204,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(21365008;61105004)

摘　　要：针对病理图像中有丝分裂核的形态多变,而难区分、难检测的问题,提出一种基于改进深度卷积神经网络的计算机辅助有丝分裂检测算法。对原病理图像利用人工标签分割出有丝分裂核和非有丝分裂核小块作为候选集;利用ZCA白化方法对其进行预处理;对处理后的图像利用改进的卷积神经网络(CNN)逐层提取数据的高维特征;利用Softmax分类器进行分类。并用GPU进行加速实验以ICPR2012大赛数据集为例,通过实验其综合评价指标F-measure达到了0. 921 5。结果表明所提算法优于传统算法,可以更好地应用到有丝分裂检测中。A computer aided mitosis detection algorithm based on improved deep convolution neural network is proposed to solve the problem that the morphology of mitotic nuclei in pathological images is changeable and difficult to detect. The nuclei mitotic and non-mitotic nuclei of original pathological images were segmented by artificial tags as candidate sets. ZCA whitening method was used to preprocess the images. The improved convolution neural network （CNN） was used to extract the high-dimensional features of the data after processing the images. The Softmax classifier was used for classification. Using GPU to accelerate, taking the ICPR2012 contest data set as an example, through the experiment, the comprehensive evaluation index F-measure reached 0.9215. The results show that the proposed algorithm is superior to the traditional algorithm and can be better applied to mitosis detection.

关 键 词：ZCA白化 深度卷积神经网络 GPU 病理图像 有丝分裂检测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]