Computer Vision and Machine Learning Approach for Malaria Diagnosis in Thin Blood Smears from Microscopic Blood Images

Abstract

Malaria is still a serious health problem globally, which is caused by a parasite of genus Plasmodium through Anopheles mosquitoes. An automated and accurate diagnosis is needed for an appropriate intervention to reduce mortality and to prevent anti-malaria resistance. In general, pathologists investigate blood-stained slides through conventional microscopy for diagnosis. However, these approaches are done by a clinical expert which biased, error-prone and moderate. This research addresses the development of robust computer-assisted malaria diagnosis in light microscopic blood images. In addition, microscopic images were obtained through stained slides which consist of illuminations and noise levels. To overcome this situation used computer vision approach i.e., improved k-SVD denoising method, fuzzy type II-based segmentation, feature extraction through local and global features, feature selection with Extra Trees Classifier and classify the different stages of malaria have been explored. The proposed classification strategy can be achieved by unifying Extremely Randomized Trees (ERT). The obtained experimental results showed Extremely Randomized Trees classifier accuracy of 98.02% that have been achieved in the process of malaria diagnosis.