采用融合ResNet和Transformer的U-Net进行疟疾感染红细胞分割  

Segmentation of malaria-infected erythrocytes using U-Net incorporating Transformer and ResNet

作　　者：刘潇霜 张伟[1] LIU Xiaoshuang;ZHANG Wei(School of Information Engineering,Gansu University of Chinese Medicine,Lanzhou 730000,China)

机构地区：[1]甘肃中医药大学信息工程学院,甘肃兰州730000

出　　处：《中国医学物理学杂志》2024年第2期191-197,共7页Chinese Journal of Medical Physics

基　　金：甘肃省教育厅创新基金(2022B-113)。

摘　　要：针对疟疾感染红细胞图像分割模型分割性能不高的问题,提出一种改进的U-Net网络模型,融合ResNet和Transformer。首先编码器部分使用ResNet,加深特征提取网络,以提取更深层次的特征;然后将ResNet输出传入Transformer模块进行目标区域特征的加强;最后通过解码器模块进行特征融合并输出结果。在疟疾显微图像数据集上,本文方法的Dice相似系数、平均交并比、类别平均像素准确率均优于U-Net网络,分别达到了87.40%、76.85%、85.28%。本文方法可以提高疟疾感染红细胞图像的分割精度,为疟疾诊断提供更有效和准确的解决方案。A novel U-Net network model which integrates ResNet and Transformer is proposed to address the problem of poor malaria-in fected erythrocyte performance of the existing models.ResNet is used in the encoder to deepen the feature extraction network for extracting deeper features,and the ResNet output is inputted into Transformer module for the feature enhancement in the target area,and finally the decoder module is used to perform feature fusion and output the results.The experiment on the malaria microscopy image dataset shows that the proposed method outperforms U-Net in Dice similarity coefficient,mean intersection over union,and mean pixel accuracy,reaching 87.40%,76.85%,and 85.28%,respectively.The proposed method can improve the accuracy of malaria-infected erythrocyte segmentation and provide a more effective and accurate solution for malaria diagnosis.

关 键 词：疟疾 U-Net TRANSFORMER 语义分割 

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