基于卷积神经网络的肾小球病理图像分类算法  

Glomerular pathological image classification algorithm based on convolutional neural network

作　　者：孙晓晗 孔祥勇[1] 吴滢[2] 王平 蔡健 彭瑞阳 王钰泽 SUN Xiaohan;KONG Xiangyong;WU Ying;WANG Ping;CAI Jian;PENG Ruiyang;WANG Yuze(School of Health Science and Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Department of Pathology,Children's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine,Shanghai 200040,China;Department of Nephrology and Rheumatology,Children's Hospital Affiliated to Shanghai Jiao Tong University School of Medicine,Shanghai 200040,China)

机构地区：[1]上海理工大学健康科学与工程学院,上海200093 [2]上海交通大学医学院附属儿童医院病理科,上海200040 [3]上海交通大学医学院附属儿童医院肾脏风湿免疫科,上海200040

出　　处：《中国医学物理学杂志》2022年第10期1313-1320,共8页Chinese Journal of Medical Physics

基　　金：国家自然科学基金青年科学基金(61906121)。

摘　　要：病理切片中肾小球自动分类是诊断肾脏病变程度和病变类型的关键。为解决肾小球分类问题,设计了一个基于卷积神经网络的完整肾小球分类框架,选用SE-Resnet作为图像分类模型,将原有模块中卷积层改为参数量更小的卷积块,在保证网络性能的前提下减少网络参数。实验结果表明,相比于其他分类算法,该算法表现最优,在肾小球系膜细胞增生、肾小球新月体形成、肾小球局灶性节段性硬化、正常肾小球的分类任务中达到了96.93%的准确率,说明该分类算法能够较好地对肾小球病变进行识别。The automatic classification of glomeruli in pathological sections is the key to diagnose the degree and type of renal lesions. A complete glomerular classification framework based on convolutional neural network is designed for realizing glomerular classification. SE-Resnet is used as the image classification model to change the convolutional layer in the original module into convolution blocks with a smaller number of parameters, thereby reducing network parameters on the premise of ensuring network performance. The experimental results show that compared with other classification algorithms, the proposed algorithm has the optimal performance and achieves 96.93% accuracy in the classification of mesangial proliferative glomerulonephritis, crescentic glomerulonephritis, focal segmental glomerulosclerosis and normal glomeruli, indicating that the classification algorithm can better identify glomerular lesions.

关 键 词：卷积神经网络 SE-ResNet 图像分类 肾小球 病理图像 

分 类 号：R318[医药卫生—生物医学工程] TP391.7[医药卫生—基础医学]