Malaria Parasite Recognition in Thin Blood Smear Images using Squeeze and Excitation Networks

Abstract

Malaria is a blood disease that is caused by the Plasmodium parasites transmitted through the bite of female Anopheles mosquito. To detect the existence of this parasite, the experts usually examine the thin blood smears. However, this requires considerable expertise in order to precisely make a distinction between the two categories. The result is that methods fail when the task of classification is largely-scaled. In recent times, researchers have started using machine learning techniques which require careful analysis of morphological, textural, and positional variations of the region of interest (ROI) in order to extract hand-engineered features. In this study, we tend to present an advanced method to automate the above process based upon the pre-trained CNN based model, SeNet. This model acts as a feature extractor towards classifying parasitized and uninfected cells. To carry out the process, we have used the dataset of microscopic images of red blood cells provided by United States National Library of Medicine. Our results show that the model achieved an accuracy of 97.24 % in identifying the malarial parasite in red blood cells. The AUC/ROC score (Area under Curve) came out to be around 0.97. The training loss was calculated using Categorical Cross-Entropy which was around 0.075. Statistical validation of the outcomes reveals the use of pre-trained CNNs as a favorable tool for feature extraction for this purpose.