Semantic Segmentation of the Growth Stages of Plasmodium Parasites using Convolutional Neural Networks

A child under five dies of malaria every two minutes. Limited access to diagnosis and treatment are some of reasons for this large death toll. Currently, the standard method of diagnosing malaria is manual microscopy which is expensive, tedious and prone to errors. This research proposes semantic segmentation of the growth stages of the Plasmodium parasites to aid effective diagnosis. By employing data augmentation and transfer learning, the proposed method achieved an accuracy of 85.86% on dense predictions of trophozoites, gametocytes and normal cells using 353 Giemsa-stained thin blood smear images.