Algorithms used in facial recognition improve malaria diagnosis
A new paper published on the Public Library of Science details a new, innovative malaria diagnostic tool based on computer vision algorithms comparable to those used in facial recognition systems, according to a report by Science 2.0.
With more than 200 million new malaria cases annually, high-quality microscopy – which requires well-trained staff and can be incredibly time-consuming — remains the most accurate method for detecting malaria infection. In an effort to increase throughput in diagnotics and make it less labour intensive, researchers proposed and evaluated a diagnostic aid based on a computer vision algorithm that analyzes an average of more than 50,000 red blood cells in a thin blood film, ranks sample areas according to probability of infection and presents a small subset of the most likely infected cells as a single panel to the user. The final diagnosis is done by a health-care professional based on the visualized images.
Using this set of diagnosed samples, the researchers proved that their method was as accurate as the quality criteria defined by the World Health Organization.
The study showed that in the test setting, more than 90% of the infected samples were accurately diagnosed based on the panel, while the “flawed” samples were deemed low quality and in an actual diagnostic setting would have led to further analyses.