Tens of thousands of times a year, a technician places a drop of blood on a slide and peers at it under a microscope, searching for malaria parasites. Making a definitive diagnosis requires the technician to look at up to 300 different fields of view over roughly half an hour. This process is repeated over and over, day after day, on every continent except Antarctica. It’s tedious work, but it saves lives. Malaria parasites infect over 200 million people and kill 400,000 every year, mostly children in Africa.
Trained and experienced malaria microscopists are rare, however. Fewer than 100 people in the world have the World Health Organization’s highest certification, Level 1. “At very low infection levels, finding malaria parasites in a blood sample is the equivalent of finding a handful of marbles in a football field in about 20 minutes,” says Cary Champlin, an electrical engineer with the Intellectual Ventures Lab in Bellevue, Washington. He and his research team believe the process can be automated. Computers can learn to recognize faces and fingerprints—why not malaria parasites?
Usually a creative technical solution like this runs up against the unforgiving economics of health care in the developing world: There’s so little money to be made that commercial companies don’t have an incentive to invest in solutions.
But IVL is not your average research lab. Founded by Nathan Myhrvold, formerly chief technology officer at Microsoft, IVL is a spinoff of his company Intellectual Ventures, which develops and licenses technology patents. Since its founding in 2007, IVL has become one of the most diverse and advanced invention labs in the world. Its non-descript office park building overlooking I-90 is full of enough tools and toys to give a mad scientist an aneurysm, from 5-axis milling machines to 3-D printers to a rocket engine. Among many envelope-pushing projects, it’s pursuing new ways of both diagnosing and treating malaria, with the lofty goal of eradicating the parasite from Earth.
“The tech industry has almost entirely been about making tools and toys for rich people,” Myhrvold says. “We sat down and thought, what are some of the ways we could direct technology to the poorest people on Earth—to transform the lives of people who need their lives transformed?”
With financial backing from Myhrvold and partners including the Gates Foundation Trust, IVL’s 140-odd scientists and researchers have the freedom to focus on what they call “atypical approaches to very large problems.” In other words, pursuing ideas that others might consider too advanced, too out there, or over too long a timeframe to be practical or financially feasible—ideas like finding safe ways to generate nuclear power, hacking the atmosphere to reduce global warming, or ridding the world of one of the deadliest diseases in history.
By some estimates, malaria has killed more people than any other parasitic disease. Over 90 percent of malaria deaths are caused by the Plasmodium falciparum protozoan, which makes red blood cells stick together inside small blood vessels. Most cases can be treated if they’re caught early. But the parasite reproduces quickly, and infected people often don’t show symptoms until it’s too late. Being able to find the parasite in a blood sample is often a matter of life and death.
Champlin and his team set out to create a cheap, portable tool that can do the job quickly, accurately, and—the most revolutionary part—automatically. The result, called the Autoscope, is a standard clinical microscope with a brain in the form of a laptop computer running custom image-recognition software. The central algorithm is based on what’s called a convolutional neural network: a highly simplified software model of an animal’s visual cortex.
The lab trained the system using four million scanned images of infected and non-infected blood samples, one of the largest databases of its kind in the world. Developing the algorithm took a lot of computing power: Early training phases could involve close to a trillion calculations over three days.
Identifying malaria parasites means spotting certain visual cues, like size: A Plasmodium falciparum parasite is three microns in diameter, compared to six to eight microns for a human red blood cell. But every parasite looks slightly different, and blood samples also contain things like dust particles, platelets, and stains. And teaching a computer to mimic the human brain is hard when you don’t fully understand how the brain works.
“First we tried giving the system our own parameters—what we think we’re looking for—like ’round,'” says Champlin, who led the project. “But that’s based on our perspective—and that’s the limitation.” When you give a computer a set of rules to follow, it’s going to be limited by those rules, unable to learn, adapt, and infer on its own. (And we’re not always even consciously aware of our own rules.)
When the human-directed approach wasn’t efficient enough, IVL let the software decide what the most significant features were. Certain factors it chose were obvious, like edges and colors. “But some of it, nobody fully understands why they’re important,” Champlin says.
What mattered was that it worked. The software eventually taught itself to diagnose malaria with the accuracy of a Level 1 expert. It can even identify the two major malaria species and give an overall parasite count. It works by creating a thumbnail gallery of the red blood cell images it considers to have the highest probability of carrying malaria, which a human microscopist then confirms. “This ‘assisted diagnosis’ approach removes the tedious searching part,” Champlin says. “It turns a 30-minute job into a 30-second job.”
Seven custom-built Autoscope prototypes are currently being tested in the field around the world. And IVL is also looking into using the system to detect other blood-borne diseases like sleeping sickness and leukemia.
In addition to the Autoscope, IVL is pursuing other bold strategies for diagnosing malaria in the field. In a conference room near the insectory, where mosquitoes are raised for testing, Kevin Nichols, an analytical chemist, holds up a plastic device that looks like a flat, oversized thumb drive. It’s not that impressive, but if the early test results aren’t a fluke, this gizmo could be another game-changer in the war against malaria. It’s called a lateral flow assay (LFA) test, a quick and easy way to detect malaria parasites in a blood sample—but this one is special.
One of the biggest hurdles to eradicating malaria is how stealthy it is. Infected people often don’t show any symptoms of the disease or detectable levels of parasites or antigens in their blood. As many as 60 percent of all malaria cases go undiagnosed; one study in Senegal suggested over 90 percent of exposed people could be infected.
Asymptomatic malaria is a chronic and transmissible condition that can affect everything from children’s cognitive function to maternal mortality. “This is one of the Achilles’ heels of the whole fight,” says David Bell, who is coordinating IVL’s malaria efforts. This is especially true in places where the disease has almost been eliminated, making anyone who is still infectious like a spark that could re-ignite a fire.
LFAs have been used since the 1990s to detect everything from malaria to dengue fever to the flu. You’ll recognize the mechanism if you’ve ever used a home pregnancy kit: You put a drop of fluid at one end, then wait for a chemical reaction to occur and for results to appear in a window at the other end. It’s a simple, inexpensive test, but current low-cost malaria LFAs aren’t sensitive enough to detect low levels of antigens (that is, evidence of infection). So IVL set out to change that.
The first step was to start with a bigger sample, Nichols says. “The current test only uses five microliters of blood, much smaller than a drop on the tip of your finger.” IVL’s version of a malaria LFA uses 20 times as much blood, significantly raising the odds of detection. Capillary action draws the sample up a strip of paper, where it hits built-in delays that allow different chemical reactions to happen at different times. “We want to slow everything down,” Nichols says. “We’re trying to improve the sensitivity by adding greater control to how the chemistry takes place.”
It took about a year to get IVL’s device working, but the results so far have been enough to make the most sober researcher start sounding like a kid on Christmas. In a lab setting, the device has shown 100 times more sensitivity than current tests, Nichols says.
It’s good to remain skeptical as long as possible, Bell admits. But even if the results from ongoing field trials are much less dramatic, as they often are, the device could still make a huge difference by making it possible to diagnose asymptomatic carriers in the field, letting them know to seek treatment. “This test could truly change the approach to malaria elimination,” Bell says. The device will still have to be inexpensive—IVL is aiming for a price of $1 each—and the demands of the market dictate a thin profit margin. Even so, IVL’s unique ability to back otherwise unfundable research projects makes it more likely the device will get into doctors’ hands eventually.
“There are a million academic papers on all sorts of clever things you can do to be useful to lots of people,” Nichols says, “but almost none of them go anywhere, because who’s going to pay for it?” The typical biotechnology company puts very little money toward research and development in low-cost diagnostics, but at IVL, “We have the capability to do that, which is pretty unique. We at least have the potential to get stuff out there.”
A 100- or even 10-fold increase in sensitivity for an LFA test would be a big deal, says David Sullivan, a malaria expert at the Johns Hopkins Bloomberg School of Public Health. So would automating the process of scanning slides. “The tough thing is doing it in the field,” he says. “In the end, it’s survival of the fittest. If the device works, it’ll be used. If not, it won’t.”
To truly eliminate malaria, it’s not enough to be able to diagnose infected people. They need treatment, and even though medications work well against P. falciparum, they are frequently counterfeited and ineffective. Demand is high and fake malaria medicines can be much cheaper to produce, since, unlike the active ingredients in some other drugs, the active ingredients in malaria drugs are often the most expensive components. An estimated 7 percent of antimalarial medications on the market are fakes, says Ben Wilson, head of IVL’s machine learning group.
Counterfeits don’t just leave malaria sufferers untreated, Wilson says. They also undermine public confidence in the authentic medicines, a major problem reported by non-governmental organizations on the front lines. One of the biggest counterfeit busts ever came in 2012, when a cargo container from China was intercepted in Angola. Hidden inside a shipment of loudspeakers were 1.4 million packets of the malaria drug Coartem—all fake. “It’s not someone doing this in their backyard,” Wilson says. “Some criminal organization with a large drug factory spent two months cranking them out. Who knows how many of these batches have gotten through?”
One of the most common ways to detect counterfeits is with a near-infrared (NIR) spectrometer, which bounces light inside a pill and scans whatever reflects back out. The pattern of light wavelengths that get reflected reveals the chemical composition. These professional-grade scanners can be extremely expensive, but in recent years handheld consumer models have become available. Paired with smartphones via Bluetooth, they’re designed for things like testing how ripe fruit is. Wilson and his colleagues wondered if the same cheap consumer models could be used to distinguish real malaria drugs from fakes.
The hardware worked, but to make it detect counterfeit malaria meds IVL needed to create custom software to interpret the NIR data. It took about a year to develop, using machine learning and data from large drug-research libraries around the world. The key lies in quality control, Wilson says. “A reputable drug manufacturer will make the exact same pill every time, while counterfeiters can’t make the same pill twice. They just use whatever’s sitting around as filler.” On a graph of NIR data showing major ingredients in pills, the ingredients in real ones cluster tightly together—indicating precise manufacturing—while fakes are all over the place.
The combination of IVL software and off-the-shelf technology is similar to the Autoscope approach, and the results have been just as promising: In lab tests, the hand-held NIR spectrometer has never mistaken a counterfeit pill for a real one. Lab tests have been “as encouraging as they could possibly be,” Wilson says. “It seems to work every time. It’s a home run.” But again, the true test will be in the field, he says.
Even though detecting fake Viagra would probably be more profitable, the IVL team is focusing its NIR work on antimalarials, as well as antibiotics and antituberculosis drugs. The approach could also work on contraceptives, up to half of which are counterfeited in South America and Southeast Asia.
Another possibility is detecting substandard medications, which were either made incorrectly or stored under bad conditions. These aren’t just limited in effectiveness; they can also contribute to drug resistance—one possible factor in how the malaria parasite became resistant to chloroquine, which used to be highly effective but has lost much of its power as the parasite has adapted. This would be harder, Wilson admits, since the pills are produced in the same facilities as good drugs and therefore aren’t as chemically distinct.
Facundo Martin Fernandez, an analytical chemist at the Georgia Institute of Technology, says that detecting substandard medications would be an important litmus test for the IVL device, especially since they are more common than pure counterfeits. “NIR is well-suited for catching simple fakes,” he says, where a pill contains no active ingredients or the wrong ones. “Its main advantage is that it’s a low-cost technique.” But when you start comparing pills that have varying amounts of the same, proper ingredients, he says, NIR may need to be complemented with other techniques. “It may give you a red flag, but you may need to confirm the results.”
One promising lab study on IVL’s spectrometer was published in February, another is being drafted, and IVL is analyzing data from field trials held this past fall in Southeast Asia. The next step would be to find a commercial-production partner. “This is what we characterize as a ‘quick win’ project,” Wilson says. “We’re doing the relatively small job of making the software, and telling people who care that this technology exists.”
“I think I will live to see malaria eliminated—certainly to see it shrunk an enormous amount,” says Myhrvold, sitting in his office wearing an “I <3 Gluten” T-shirt. The room is full of a museum’s worth of objects reflecting his fascination with technology, science, and history: fossil dinosaur toenails, antique movie cameras, a table made from part of a nuclear reactor. He’s brimming with energy and knowledge, veering off on detailed tangents punctuated by high giggles.
“We know how to get rid of malaria,” he says. “We’ve done it lots of places. The trouble is it costs a lot of money, and you need a functional health-care system to do it.” That’s part of the reason IVL licenses its health-care technology for free, or close to free, in developing countries through its Global Good Fund, which invests in humanitarian projects in the high-risk initial stages of research and development. If a private company takes one of IVL’s innovations to market and can make a profit, he says, that’s great—but making money isn’t an absolute requirement. “It is kind of cool to try to save the world.”
A version of this story originally appeared in the August/September 2017 issue of Pacific Standard.
