Abstract

Info

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

Author(s):: LIU Xiaoshuang; ZHANG Wei; School of Information Engineering, Gansu University of Chinese Medicine, Lanzhou 730000, China

Abstract:: 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.