Plasmodium Parasite Detection Using Combination of Image Processing and Deep Learning Approach

The development of an intelligent system for automated malaria detection became the one of challenges since its application supported the examination process which was conducted manually by the doctor or medical personnel. Some previous studies have been done to overcome those problems. However, most of them still have problem in detecting parasite candidates. Hence, their proposed methods did not successfully detect all parasite candidates and remains a large number of false-negative. Actually, the misdetection problem occurred since the characteristic of parasites seems unclear. To overcome these problems, we applied image processing technique and deep learning architecture to detect and to ensure whether the detected candidate is a parasite or not. Our proposed method was applied to 46 digital microscopic images provided by the Department of Parasitology, Universitas Gadjah Mada and Eijkman Institute for Molecular Biology. The proposed method comprised of four steps which are normalization process using GGB (green, green, blue) color transformation, segmentation process using Otsu followed by some morphological operations, object labelling using BLOB analysis, and classification using deep learning. Our detection process successfully detected all parasites and the classification process achieved an accuracy, sensitivity, specificity, PPV and NPV of 98.97, 100, 98.08, 97.85, and 100% respectively. This result shows that our proposed method achieved outstanding performance in both detection and classification process which indicates that our proposed method had the potential to be implemented as an intelligent system for supporting the parasitologist in conducting rapid assessment of plasmodium parasite infection.