Two men standing next to each other in a lab.
Paul Lebel (left) and Chris Charlton stand in front of microscopes in the making. Photo by Yujie Zhou, Sept. 29, 2023.

With all the talk of artificial intelligence swallowing the world, a group of San Francisco scientists have finally discovered a real-world application: Detecting malaria, a deadly disease that claims more than half a million lives each year

“Every time we show it to a group in different countries, everybody wants it,” said Chris Charlton, who leads the mechanical design for the team at the Chan Zuckerberg Biohub, a nonprofit research organization funded by Mark Zuckerberg and Priscilla Chan that collaborates with universities like Berkeley, University of California, San Francisco, and Stanford. 

Charlton and his colleagues have created the so-called Remoscope, a microscope that uses machine learning to detect malaria parasites, a process currently accomplished (with varying efficiency) by human technicians. 

After training on images of parasites and healthy cells, the AI model is capable of going through 40 patients’ blood samples every day, almost doubling the output of a single technician. 

Detection is fast: A human technician collects blood from a patient, and the machine takes about 15 minutes to analyze results. For very sick patients who carry large quantities of malaria parasites, the analysis can be done in as little as a minute. Current technology takes half an hour before it can even be ready for human review. 

And it is in high demand: So far, the Remoscope has been presented to interested parties from at least 15 countries, particularly those in at-risk equatorial regions like Cambodia, India, Indonesia, Colombia and Malawi, where the disease is transmitted by mosquitoes.

The group has just flown back from Uganda, where malaria is the leading cause of death, especially among young children. Researchers conducted field tests that reached accuracies comparable to the human technicians in well-funded facilities.

Most of Uganda’s small, government-run health centers continue to use the same testing method they used a century ago, according to Paul Lebel, project lead of Remoscope. Because of the scarcity of resources and the enormous need for testing, experienced technicians usually spend 16 hours a day looking through microscopes, laboriously examining 200,000 cells on each slide and counting each malaria parasite with a handheld clicker. (Remoscope can examine two million cells.)

The Remoscope automates that process and can help avoid mistakes made by overworked technicians, who occasionally confuse platelets for parasites or miss counting parasites. Even a single mistake can be serious. “Usually, it’s children who would be at risk of falling very ill. So if you miss the infection and the child was sent home, they could die,” said Lebel.

Remoscope also “cuts out all the labor and the chemicals that [health centers] need. And it cuts out the expertise,” said Lebel. For now, the cost is around $3 per test, mainly for the testing chip, but Lebel said this would drop to $1 after future scaling. By comparison, broadly deployed rapid diagnostic tests cost between $3 and $8 for a single test.

Remoscope doesn’t require “expertise” either: After a human loads the blood onto a chip, the microscope conducts a detection test on its own, which may also enable it to be deployed to remote villages where experienced microscopists are not stationed. That would avoid paying for technicians, who must receive months of training and become certified by the World Health Organization before going on duty. 

  • An image of a computer screen showing a number of images.
  • A man writing on a whiteboard in a lab.
  • Two men holding up a piece of plastic in a lab.
  • A group of electronic devices on a table in a lab.

In Uganda, agricultural workers are the most exposed to malaria, particularly those working near bodies of water, who may suffer hundreds of mosquito bites every day. Adults are unlikely to die from malaria, so workers often choose to abstain from testing, for fear of missing work. 

But their bodies can still carry the parasites, which can be passed on to others by any mosquito who bites a malaria carrier and then bites someone else.

Lebel hopes Remoscope can reduce the time burden of testing.

“There’s probably a parent working in the field somewhere. There’s a parent with some number of children. She has to pack them up, bring them, transport them an hour or two to this clinic. So it’s basically going to take her all day to do this,” said Lebel.

The limitation of testing has been one of the major obstacles against eradicating malaria, said Lebel. In Uganda, for example, there is a “40 to 50 percent prevalence that a random person on the street has malaria,” said Lebel. But it is simply impossible to test everyone under current practices, including testing adults who are asymptomatic. 

The Remoscope project began much like other AI projects: By ingesting large amounts of data.

Joe DeRisi, president of Chan Zuckerberg Biohub and an infectious disease expert who has studied malaria for 25 years, suggested to Lebel that they should put malaria under a UV microscope which Lebel had invented. 

“Machine learning really is the only way of analyzing the images, because the images have subtle differences,” said Lebel. “The parasite can change shape, it can change appearance. There’s a life cycle. It’s changing over time.”

The 10-member team used Yolo, an existing open-source network that is generally trained to detect objects like a car or a puppy, and fed it some 100,000 images of malaria-infected cells and hundreds of thousands of images of healthy red blood cells.

Also, the AI model was sped up tenfold, so it could run efficiently, even on a low-cost computer deployed in a village medical clinic.

Lebel emphasized, however, that humans have not been entirely removed from the process. At the end of each examination, a technician will review a readout with thumbnails of the parasites that the machine learning model selected. “We’re doing our best to [make the AI] understand all the diversity of types of cell images,” said Lebel. “But, ultimately, you don’t know if there’s always some new case that comes in. You can’t predict what a machine-learning model is going to do.”

For Lebel, the upshot is clear: During the recent site visit to Uganda, the local lab manager, who “has been involved personally in testing several other instruments for doing this, even commercially produced ones, told us that he thinks our project is the defining project of his career” Lebel said, adding that he harbored significant “self-doubt” about the efficacy of the Remoscope because of the intense competition between global scientists in this field, which has diminished over the time he’s spent with those fighting malaria on a daily basis.

Still, the upshot may be years away: There are working prototypes of Remoscope, but Lebel said it could take some five years to find business partners, gain regulatory approval and deploy the devices in remote areas. 

The wait, however, may be worth it. Lebel said that malaria has claimed so many lives over so many decades, many have lost a sense of urgency.

“What struck me is that malaria has been around for so long,” he said. “People have practically forgotten. When covid came at its peak, it only briefly exceeded the malaria deaths.” 

Disclaimer: The Chan Zuckerberg Initiative is a funder of Mission Local, and Joe DeRisi serves on Mission Local’s board.

Follow Us

REPORTER. Yujie Zhou is our newest reporter and came on as an intern after graduating from Columbia University's Graduate School of Journalism. She is a full-time staff reporter as part of the Report for America program that helps put young journalists in newsrooms. Before falling in love with the Mission, Yujie covered New York City, studied politics through the “street clashes” in Hong Kong, and earned a wine-tasting certificate in two days. She’s proud to be a bilingual journalist. Follow her on Twitter @Yujie_ZZ.

Join the Conversation

1 Comment

  1. As the manage of the team that works on this project, I’d like to make it clear that the team is bigger than just Paul Lebel and Chris Charlton. The following engineers have also worked really hard on this project and deserve to be acknowledged: Michelle Khoo, Ilakkiyan Jeyakumar, and Axel Jacobsen.

    votes. Sign in to vote
Leave a comment
Please keep your comments short and civil. Do not leave multiple comments under multiple names on one article. We will zap comments that fail to adhere to these short and very easy-to-follow rules.

Your email address will not be published. Required fields are marked *