High throughput image processing for malarial blood analysis
Maikel van Herpen

One of the most dreaded diseases in the world is malaria. The disease affects more than 350 million people per year and is responsible for more than 1 million deaths yearly. Malaria is caused by a parasite of the genus Plasmodium and a human can become infected after being stung by an infected Anopheles mosquito. Eventually, Plasmodium parasite will contaminate the host’s blood, rapidly multiplying within blood cells. In principle, malaria is a curable disease, though the medicines that have to be administered differ for the four species of Plasmodium that affect humans. If the wrong medicines or doses are administered, this could lead to immunity of the Plasmodium parasite. Nowadays, Plasmodium Falciparum, the most deadly form of malaria, is immune against multiple drugs in large parts of the world. Therefore, it is important to diagnose if patients suffer from malaria quickly and accurately. This reports describes a computationally efficient method to analyse a very large amount of microscope blood cell images to automatically select the regions of the image, where Plasmodium parasites are possibly present. In a later stage, more processing effort can be assigned to these regions for further diagnosis. The final aim of the project is the construction of a portable device, that can do automatic malaria diagnosis in the field. Firstly, the biological background of malaria is presented. Secondly, a description on the earlier research on the field of malarial blood analysis is given. Consequently, it is shown that the number of erythrocytes can be estimated efficiently and with enough accuracy. Subsequently, the AdaBoost algorithm is introduced as a method to select relevant features from blood cell regions. Finally, a boosted classifiertree is trained to efficiently identify infected blood cells.