First Step of Automated Malaria Diagnosis: Evaluation of Focus Functions in Thick Blood Smear Images

Abstract

According to World Health Organization, light microscopy is the diagnostic standard of malaria. This diagnosis requires examination of both thin and thick films from the same patient. However, in most large health clinics and hospitals, the quality of microscopy-based malaria diagnosis is frequently inadequate. Automatic microscopy diagnosis allows an increase in the number of fields of view to be analyzed, providing more accurate diagnosis, while reducing the time required for that purpose. Automatic focusing of the microscope is an essential component of automated microscopy; it is the first step of automated malaria diagnosis. In this work, we implemented the “classical image-analysis-based auto-focus techniques” approach using nine-focus function in order to identify the best focus function for thick blood smear images. Because some previous works have shown that the accuracy focus functions sometimes depends on content of the processed images, and the specimen can determine which metrics should be more adequate, we proceeded two experiments. In experiment #1, we evaluated the focus functions in an image-stacks dataset (338 stacks and 5 images/stack). Then, we did experiment #2, this time, testing with patch images (fragments) containing Plasmodium vivax in its various life cycle phases (ring or immature trophozoite, ameboid trophozoite, schizont and gametocyte). The parasite dataset used contained 1713 patches. Brenner gradient focus function was the best in both experiments.