Detecting Malaria Parasites with Machine Learning

S. Jaeger
United States

Keywords: smartphone, machine learning, malaria, disease


This talk will present a smartphone application for computer-assisted malaria screening. Malaria, which is caused by parasites transmitted by infected mosquitoes, is a major health threat responsible for about 400,000 deaths per year. The disease is widespread in sub-Saharan Africa, where most of the cases occur and where malaria is a leading cause of childhood neuro-disability. Typical symptoms of malaria include fever, fatigue, headaches, and in severe cases seizures, coma, and death. Malaria is a curable disease; however, developing adequate diagnostics has been a major obstacle toward reducing its devastating impact on society and economy. Microscopy is a common way of diagnosing malaria, especially in resource-poor settings. Millions of blood smears are screened for malaria parasites under the microscope every year. This largely manual process involves counting of parasites and blood cells, which is prone to error and depends on the skill and experience of a microscopist. However, accurate counts are essential to measuring disease severity and drug-effectiveness. False negatives can lead to unnecessary prescription of antibiotics, losing valuable time and thus risking the progression into severe malaria. On the other hand, false positive decisions cause unnecessary use of anti-malaria drugs with potential side-effects. To automate this process, the U.S. National Library of Medicine is developing intelligent software that can detect infected blood cells and count parasites automatically, in collaboration with national and international partners. The software runs on a smartphone that is attached to the eyepiece of a microscope. This setup allows taking pictures of blood smears with the built-in smartphone camera and processing these pictures directly on the phone. Automated image analysis and machine learning methods can automatically detect parasites in the blood smear image and report them to the user. The goal is to develop this system into a tool that can increase the quality of malaria diagnosis and patient monitoring, and thus help to reduce diagnostic costs and workload in the field. The talk will touch on several important aspects in the research and development of the smartphone application. For example, the acquisition of training data and the machine learning methods that can learn the typical visual features of blood cells and parasites will be discussed in more detail. Another topic will be the implementation of the application, including the user interface, processing pipeline and general workflow, among others. The practical evaluation of the system and its general performance will also be addressed. Finally, an outlook will summarize the current state-of-the-art and discuss future research directions to make automated parasite counting an accepted diagnostic tool.