Applying deep-learning approaches for facilitating malaria diagnosis.

What is the issue or challenge that you are trying to address with assistance from a Fulbright Specialist? Currently, the standard diagnosis technique recommended by the World Health Organization, WHO, is the Romanowsky method, which is based on staining thin and thick blood smear samples and analyzing them with optical microscopy. This methodology is affordable, as it only requires an optical microscope equipped with a 100X magnification lens system, which is useful for developing countries where malaria is rampant. However, detecting the parasites (shapes of the blood cells and colors and morphology of parasites, mainly) is a very slow process that is prone to errors and should be assessed by a highly trained individual. Malaria undergoes several stages of development, and the blood samples differ depending on the type of parasite (Plasmodium (P.) vivax, P. falciparum, P. malariae, P. ovale and P. knowlesi). Each specie has a characteristic morphology and size, which depends too on the whether. In fact, some of the scientists who participate in this proposal (Dr. Liliana Jazmin Cortes Cortés, Colombian National Institute of Health, INS) are precisely dedicated to training individuals around Colombia to be able to recognize the subtle aspect in the infected samples. Due to the quality of the images collected differ from area to area and assessing whether or not a blood cell is oval (one of the indicators of the disease), we wish to use modern computational techniques to make the diagnosis not only more accurate but also faster. We propose the use of the deep-learning techniques developed by the Data NanoAnalytics, DNA, group at Center for Nanophase Materials Sciences (CNMS), at Oak Ridge National Laboratory (ORNL), in order to develop an alternative diagnosis method. Among the institutions that the Fulbright candidate will visit is the INS in Colombia, were the standard Romanowsky method has been successfully used for many years in the country. The goal of this visit is to explore whether using Deep Learning, DL, techniques can aid the manual approach. Several groups around the world have already demonstrated that DL techniques can speed up malaria detection, and even an application for iPhone has been developed for this purpose. Our objective is to determine, also, the growth stage of the malaria parasite (shape and color) because doing so will improve treatment. Both, Thin Films & Nanophotonics Group (TF&NPG) and the Bioengineering group of Pontificia Universidad Javeriana (PUJ), have a database of approximately 2000 images (1650 x 200 pixels) taken from the blood samples of patients who have had the disease. The images were obtained in different parts of Colombia and they don¿t have any Personal Identifiable Information (PII). The complete set of labeled images will be shared with the DNA group at CNMS, who will train a DCNN and a rotational invariant VAE to search for granulations and oval shapes in the complete set of images. The candidate, in close collaboration with scientists and students at hosting university (PUJ), will explore whether rVAEs can be used to classify parasites according to their shape and/or color.