Engineers at Stanford University have developed a high-efficiency, battery/solar-operated, autonomous microscope with integrated artificial intelligence that automatically diagnoses malaria in blood smears - a previously tedious process done manually, slide-by-slide, by technicians in the field.
The researchers call it Octopi, and believe it could save countless lives through earlier and more accurate diagnosis - and perhaps someday lead to outright eradication of the parasites that cause malaria, the world's deadliest infectious disease.
"Currently, a human sits at a microscope looking at slides, hour by hour, counting the infected cells by hand. Each sample takes half an hour. A technician can do maybe 25 people a day - and that's a 12-hour day," lays out Manu Prakash, an associate professor of bioengineering in the Schools of Engineering and Medicine at Stanford University and Octopi's inventor. "For the first time, Octopi can now do an accurate diagnosis in minutes in the middle of nowhere with no other infrastructure. No power? No internet? No problem!"
Deadly but defeatable
Malaria kills 600,000 people each year, mostly children and mostly in the under-resourced countries of Central Africa. Millions more are among the walking infected, unwittingly passing the disease to others through mosquito bites. Faster, more accurate diagnosis would not only improve treatment but identify asymptomatic infections to help control spread of the disease.
Building a better blood smear
Alongside Octopi, Prakash and team have made a complementary invention to automate and standardize the process of preparing the blood samples on microscope-ready glass slides, called a smear.
"We had hit a roadblock a couple of years ago," Prakash recalls. "We realized results were inconsistent based on who was doing the blood smear. In 200 years of slide-making, no one had taken the time to understand what makes a smear 'perfect'. So, we had a new engineering challenge on our hands."
They created Inkwell, a passive mechanism to make perfect slides every time without electricity. It uses a clever capillary mechanism to ensure each blood smear contains a perfect thin layer of 20 million blood cells ready for imaging.
Inkwell is open source and can be printed on a commercial 3D printer using less than $5 in parts, requires no electricity, and has already been tested and used in more than 15 countries worldwide.
Octopi is remarkably efficient and highly sensitive. It can scan 1 million blood cells per minute - a 100-fold increase in efficiency. And the tool is so sensitive it can spot concentrations as few as 12 infected cells in microliter sample of blood with upward of 5 million cells with near-100-percent specificity.
"Octopi is both fast and precise - but also quantitative," Prakash said, recalling a time he was in a rural clinic in India where his collaborators were worried that a certain child had cerebral malaria. The team urgently needed to know not only if the child had malaria, but what their exact parasite count was. Octopi was able to deliver that count.
"Octopi can detect small infection loads and quickly calculate severity levels that are key to early treatment, establishing drug regimens, and to saving lives," Prakash said, emphasizing that public health providers have just two weeks to spot a new infection before malaria turns deadly.
Designed for affordability
To keep costs down, Octopi favors inexpensive optics over costly glass lenses necessary for modern microscopes. Affordability was a top-of-mind concern for Prakash and lead graduate student, Hongquan Li, in designing Octopi. Hongquan and Prakash first started the project nearly a decade ago, anticipating the falling costs of computation and the rise of new machine learning frameworks.
Prakash lab
An ad hoc malaria screening post in rural Tanzania.
A community health worker uses the Octopi 2.0 tool for malaria screening in an evaluation setting at a rural school.
The tool in an evaluation setting for screening malaria in Tanzania.
Training session in a hospital in Monrovia, Liberia.
Previous
Next
Close
After nearly 10 years of refinement and testing these tools in nine countries across Africa with hundreds of collaborators, their bet finally paid off. Octopi is expected to cost $1,000 per unit, whereas current robotic microscopes can cost $100,000 per unit, or more. Further, Octopi is powered by rechargeable batteries or solar panels to help it function in remote, off-grid regions of the world where malaria thrives.
The team circumvents the need for expensive hardware by training Octopi to look for a simple spectral shift that occurs when infected blood cells are illuminated under ultraviolet light. Hongquan and Prakash borrowed the concept from astronomy, where the exact chemistry of a faraway star can be revealed with a low-power optics from Earth.
"This spectral shift can be recognized easily with lower-cost optics," Hongquan explains. "The infected cells light up, and AI can spot and count them quickly to calculate disease load.
App store model
Most promising among Octopi's features, however, is its open software architecture that can be modified and customized by users around the world, Prakash said. His lab demonstrated this approach through partnership with hospitals in Nepal, where other users were able to train the same instrument to detect all four types of sickle cell anemia without modifying the hardware.
"This was an eye-opening moment for the team," said Pranav Shrestha, a key team member who also grew up in Nepal seeing the deadly toll of malaria and sickle cell anemia. The team then repeated its feat by reconfiguring the tool to identify tuberculosis.
Prakash believes that Octopi could prove to be a pivotal technology for disease diagnostics in rural settings. "By expanding the number of diseases we can screen for with the same infrastructure, we are hoping to build a universal diagnostics platform," Prakash said. "Not only can we configure Octopi to spot malaria, tuberculosis, sickle cell anemia, but hundreds of other diseases, including sexually transmitted infections, parasites like leishmaniasis and schistosomiasis, and more."
In theory, any disease that can be identified with a microscope could join the universal app store for diagnostics. The team is now building advanced annotation tools that allow any health care worker in the world to train new models. It would give rise to literally thousands of models.
"We want to create an app store for the world's most deadly diseases," Prakash said.
Researchers discuss the Octopi microscope's autonomous capabilities for rapid malaria diagnosis.
Close
Demonstrated need
There are 260 million malaria cases reported annually, a number Prakash said could reach half a billion if diagnostics and interventions do not keep pace. There are rapid antigen detection (RDT) methods available, like those used for COVID, but the parasites have evolved ways to avoid detection and RDTs are not quantitative, so they don't describe disease load.
To scale their efforts, the team is currently raising funds to launch ODION, which stands for "Open Diagnostic Imaging Observatory Network (ODION),"a first-of-its-kind research and education initiative that empowers clinicians, health care workers, entrepreneurs across the global south to collect data, train, and adapt their own diagnostic models and build an ecosystem of apps for Octopi.
An army of Octopi fanning out across the world, Prakash said, could lead to a dramatic reduction in malaria cases and deaths, and put eradication within reach.
"A thousand Octopi could image something of the order of 50 million slides per year," Prakash calculates. "If we had 10,000 of them, imagine the good we could do."
