Machine Learning Algorithms Are Now Detecting Malaria

July 13, 2016 by Ebony Calloway

Using deep learning algorithms, EE students at the University of Washington have developed a portable microscope capable of detecting and assessing malaria infections in the field.

Using deep learning algorithms, EE students at the University of Washington have developed a portable microscope capable of detecting and assessing malaria infections in the field.

The device could be a major stride in diagnosing malaria, which affects over 200 million people annually. 

Malaria is a parasitic infection most commonly spread by mosquitos. It can be detected by assessing a patient's blood sample via microscope.


Malaria as seen through a microscope. Image courtesy of the CDC.


Usually, a trained professional must be present to diagnose malaria, specifically a microscopist who can identify the malaria parasites in blood samples. But in the poorest areas of the world, where malaria is so prevalent, these professionals are in short supply.

The UW students hope to mitigate this problem with their portable malaria diagnosis device, a machine they have dubbed the Autoscope.


The Autoscope with casing removed. Image courtesy of the MIT Technology Review. 


Built for Field Work

The Autoscope is small, 15 inches tall and 7 inches wide, which makes it portable for the field. It was built to withstand up to 35 degrees celsius and 100% humidity, although extreme weather tests have not been done as of yet.

Although the microscope will cost about 1-3 times as much as microscopes currently being used in the field, the Autoscope provides more information than a microscope can and does not require a microscopist to be present.

The device scans the patient's blood sample and returns a report with a diagnosis and other details for a professional to review. This means that a technician with comparatively little training can operate the machine in these rural areas. The machine's diagnostic capabilities allow for on-site accuracy without necessarily needing to send the test results to an off-site professional.

Other recent medical devices, such as the Cardiopad, are engineered to send results to doctors and nurses in hospitals far from their patients in remote areas. The Autoscope, however, aims to provide diagnostic information immediately. 


Graphical representation of the components in the Autoscope. Image courtesy of IEEE Xplore.


Deep Learning Algorithms

The Autoscope uses deep learning, computer vision, and deep neural network techniques to provide an accurate diagnosis. The automated microscope is 90% accurate at detecting and counting the number of malaria parasites.

Deep learning algorithms and neural networks help the microscope to detect abnormal patterns in the input samples. The Autoscope detects whether or not the sample contains malaria parasites and the type of parasites present by looking at the shape, color, and texture of the objects in the image. But the Autoscope generates a report that not only detects malaria but also gives details about the malaria, like whether it is drug resistant or the density of the parasites. 

The basis of the machine learning algorithms in the Autoscope is image analysis. There is preprocessing of the sample to reduce color differences between images, then detection attempts to find the parasites in the field of view.


Images showing the Autoscope's field-of-view (left) and P. falciparum, a species of parasite that causes malaria (right). Image courtesy of IEEE Xplore.


In the detection step, ground truth images are used to train the algorithm. The output is a labeled image of parasites or non-parasites as the case may be.

A segmentation is performed on the parasites to determine where the nucleus and cytoplasm are, which is important for the classification step.

Feature extraction is then used to determine the relationship between the nucleus and cytoplasm, then a classifier score is assigned to each object in the image to determine whether it is a parasite or not.

The last step is diagnosis which makes the decision if the sample has malaria or not.


An example of the Autoscope's output, showing "parasite suspects". Image courtesy of IEEE Xplore.


The Autoscope could become an invaluable resource in helping to combat a disease that affects millions of people every day. Also, it provides useful information so that trained professionals can make more informed decisions about a patient.

Currently, the Autoscope has been tested with malaria patients but the team has plans to extend the Autoscope's diagnosing capabilities to other diseases as well. The computer vision and classification algorithms that it uses to diagnose malaria will likely yield similar diagnostic results for many different infectious diseases.

Read the team's academic paper as published for the 2015 Global Humanitarian Technology Conference here.