Malaria is pretty easy to detect inside a hospital laboratory, but the disease is prevalent in poor areas of the world where clinical access is often limited. Duke University researchers have now developed a fully automated system that can be used in the field to test for malaria with only a blood prick. Currently most testing is done using standard microscopy, and the process of staining, preparing, and visualizing the cells can be time consuming. The new Duke technique can potentially screen thousands of cells per minute and maybe screen entire villages in about a day.
The system relies on quantitative phase spectroscopy in which a laser quickly changes its color across the visible spectrum as it illuminates a cell. A sensor detects how the cell affects the incoming light at various frequencies and the data is compiled together to create a holographic image.
To actually identify which cells are infected, a deep learning algorithm that notices correlations was fed more than 1,000 examples of both healthy and infected cells. It identified a series of correlations between certain parameters of the cells and different stages of malarial infection as seen in the holograms. Testing the system on hundreds of cells resulted in an accuracy between 97% and 100%
Four cells in different stages of infection from a malarial parasite as analyzed by a new algorithm (the first image in the post is of cells under a microscope in the same stages). The algorithm uses various measures of the cell’s physical characteristics to determine whether or not it is infected.
Study in PLOS ONE: Automated Detection of P. falciparum Using Machine Learning Algorithms with Quantitative Phase Images of Unstained Cells…