Malaria Classification Using Convolutional Neural Network: A Review

Abstract

The Convolutional Neural Networks (CNNs) have been used to classify malaria parasites from blood smear images automatically and successfully gave a good result, thus enabling fast diagnoses and saving the patient. This study presents a review of the existing CNN techniques used for malaria diagnosis, focusing on the architectures, data preparation, preprocessing, and classification. Furthermore, this study discusses why the comparability of the presented methods becomes difficult and which challenges must be overcome in the future. First, we review the current CNN approaches used for malaria classification from existing research articles. Next, the performance and properties of proposed CNN approaches are summarized and discussed. The use of CNN as a feature extractor shows better performance than transfer learning and learning from scratch approaches. Unfortunately, some research uses private datasets for training and testing the proposed model. Thus it is not easy to compare with the other methods. The use of CNN in malaria diagnosis is also still limited to binary classification, namely the normal and malaria-infected erythrocyte class. Future research should use available benchmark public datasets to allow the proposed CNN method comparability and proposed a CNN model for multi-class classification such as species and life stages of malaria-causing plasmodium.