Automated Classification of Malaria Parasite Stages Using Convolutional Neural Network-Classification of Life-cycle Stages of Malaria Parasites

Malarial is a mosquito born deadly disease that quickly grows from person to person because of the infectious mosquito bite. Knowing accurately the life-cycle stages of malaria parasite is critical for accurate drag selection for early recovery. When the infected mosquito bites the host, cell morphology and appearance greatly change through four major developmental stages namely ring, trophozoite, schizont, and gametocytes in the host's liver and later in the red blood cells (RBCs). Microscopy images carry the signatures of the above changes. However, widely used image analysis based computational techniques require expertise in analyzing morphological, texture, and color variations in the images. In this study, we investigate the strength of convolutional neural network (CNN) towards effective classification of malaria parasite stages. We design a customized CNN model to discriminate five classes including the control and four malaria parasite stages as mentioned above. With an imbalanced dataset having 46,973 single-cell thin blood smear images, the proposed method achieves 97.7% average accuracy, which is about 8 similar to 10% higher when compared with a pre-trained CNN model and a widely used hand crafted feature based model using support vector machine (SVM) classifier.