Abstract

Info

Title:: Image recognition of malaria-infected erythrocytes based on graph convolutional network

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

Keywords:: malaria-infected erythrocyte; image recognition; graph convolutional network; deep learning; medical image processing

Abstract:: Objective To apply the image recognition method based on distance graph convolutional network to the image processing of malaria-infected erythrocytes for realizing the multi-stage recognition of malaria and improving the diagnostic efficiency of malaria. Methods A multi-stage malaria recognition model based on distance graph convolutional network was proposed. A radial basis function was firstly added in KNN graph construction algorithm to construct adjacency matrix and assign weights to the nearest-neighbor nodes according to the similarity between nodes, so as to weaken the effects of the distant nearest-neighbor nodes on the central node. Then, attention mechanism was introduced to update adjacency matrix dynamically in the graph convolutional network for making the model pay attention to near-neighbor nodes with higher similarity, and finally the multi-stage image recognition of malaria-infected erythrocytes was completed. Results Validated on the Malaria-MIT dataset, the experimental results show that compared with original model, the proposed method improved accuracy, precision, recall rate and F1-score to 96.18%, 96.23%, 96.18% and 96.18%, respectively. Conclusion The proposed approach can effectively accomplish the task of multi-stage image recognition of malaria-infected erythrocytes.