Bringing machine learning to last mile health challenges

SAN FRANCISCO — A new microscope will use image recognition software and machine learning technology to identify and count malaria parasites in a blood smear. The EasyScan GO, announced at MEDICA, the medical industry’s leading trade fair, is the result of a partnership between the Global Good Fund, a Seattle-based group funded by philanthropist Bill Gates, and Motic, a China-based company that specializes in manufacturing microscopes. Field tests have demonstrated that the machine learning algorithm is as reliable as an expert microscopist in fighting the spread of drug resistant malaria.

EasyScan GO is the latest example of Global Good’s partnering to bring emerging technologies to health systems in low resource settings. Based at the invention company Intellectual Ventures, Global Good is focused on developing and deploying technologies for the poorest parts of the world. It partners with others to bring the benefits of technological innovation to the hardest markets in the world to reach.

The Global Good motto is “invention saving lives.” But invention means nothing if these products are developed without any consideration of the applicability of that technology in the markets they are meant to serve, said Maurizio Vecchione, senior vice president at Global Good. He emphasized that his team cannot fulfill its mission unless that technology is affordable, appropriate, accessible, and those 3 A’s, as he calls them, are at the center of every Global Good partnership.

Proponents see enormous potential for this technology. Today, machine learning can help tell us we are sick; emerging technologies like artificial intelligence will allow us to prevent illness, according to AnthroTronix founder Corinna Lathan, who wrote a World Economic Forum blog post on ways artificial intelligence and robotics will transform health care.

Several recent partnerships point to how Global Good is working to make sure the benefits of AI, machine learning, and deep learning extend to developing countries.

The world’s first platform for infant biometrics

One in four children in the world are not registered at birth. Because they lack formal identification, it is very difficult to track their health records, including vaccinations.

Earlier this month, start-up Element Inc. announced a partnership with Global Good to develop a smartphone based platform to verify the identity of infants and children. Element Inc. focuses on developing mobile software for biometric recognition. Their aim is to deliver identity for everyone, and particularly the 1 billion people who lack proper IDs in Asia and Africa.

In Bangladesh and Cambodia, the BioNIC partnership will extend Element’s adult identity platform to children under age five, testing on fingerprints, ears and feet, irises, and more offline and on any mobile device.

“Accurately identifying patients, and linking them to their electronic medical records — which store past medical conditions, drug allergy information, medications and vaccination history — is critical to health care,” David Bell, director of the global health technologies portfolio at Global Good, said in a press release. “While adult biometrics are widely used in financial services, these have not been translated to the health sector where the need to identify a person from birth to old age presents additional challenges; namely, that infants have delicate, rapidly changing features that are difficult to capture.”

This technology has only become possible in recent years, Element co-founder and CEO Adam Perold told Devex via email.

“Mobile devices now proliferate the world, at nearly three billion units; and each year, they get better and less expensive. At the same time, deep learning — through improved algorithms and increasingly powerful computers — was driving breakthroughs in how systems process video and images. We were one of the first to apply deep learning methods on mobile devices, and as a result, we have built a platform that traverses many challenges that still prevent widespread adoption of biometrics across low-resource settings,” he said.

When problems can be tackled by deep learning, the barriers in delivery are often structural, Perold continued, explaining that it is a resource heavy approach requiring lots of data, computing power, and specialized training.

“Deep learning solves key issues of access in biometric recognition. The strengths of the approach, where algorithms train themselves directly from the data, means that anyone can be recognized regardless of their features,” he said, adding that the technology “can be deployed to continuously adapt to any changes as a person develops.”

The BioNIC partnership has the ultimate goal of creating a biometric solution capable of following patients from birth to old age, said Perold.

A breakthrough AI-powered microscope

Malaria kills almost half a million people every year, and one of the major challenges to eradication is the rapid spread of a multidrug-resistant strain in parts of Southeast Asia.

Currently, only analysis by a World Health Organization certified microscopist can accurately detect severe and drug-resistant cases. EasyScan GO automates that process with an intelligent microscope, simplifying and standardizing malaria detection in under-resourced countries where there is often a lack of trained personnel.

“Malaria is one of the hardest diseases to identify on a microscope slide,” Bell of Global Good said in a press release. “By putting machine learning-enabled microscopes in the hands of laboratory technicians, we can overcome two major barriers to combating the mutating parasite — improving diagnosis in case management and standardizing detection across geographies and time.”

The EasyScan GO uses AI-enabled software to automatically and accurately identify and count malaria parasites in a blood smear in as little as 20 minutes, Sebastian Nunnendorf, head of research and development at Motic China, said in an email to Devex. “Human capacity issues have long been a major bottleneck in low-resource settings, and the fact that we are able to address these issues through technological innovation is truly exciting,” he said.

Any new medical device will need time in the market to validate, Nunnendorf said, and like the BioNIC partnership, EasyScan GO is still early in its journey. But the company hopes to go beyond diagnosing malaria and a few other parasites commonly found on a blood film, including Chagas disease, microfilaria, and sickle cell. Success with the most difficult-to-identify disease, malaria, will pave the way for EasyScan GO to excel at almost any microscopy task, Nunnendorf said.

“AI and digital health are increasingly becoming the most exciting trend in healthcare, and with our product leadership and Global Good’s commitment to inventing for humanitarian impact, our innovations will be truly available everywhere,” he said.

Bringing AI to all 

Machine learning is the ultimate way to go faster, said Peter Norvig, director of research at Google. But speed can also lead to accidents, he told the Train AI conference in San Francisco.

“In every industry, there’s a place where AI can make things better,” Norvig told Devex. “Look at all of the AI technologies, and all problems, and it’s just a question of fitting them together and figuring out, ‘What’s the right technological match and what’s the right policy match?’”

The most obvious cases for AI use include anything a typical human can do in less than a second of thought, Andrew Ng, one of the leading thinkers on artificial intelligence, told Devex at an event at Stanford University on the role of AI in achieving the SDGs. Image recognition is one common example, but AI also has powerful applications in education and health care in both the developed and developing world, said Ng. He finds it hard to name a major industry he doesn’t think AI can transform in the next several years.

“Deep learning has driven unprecedented breakthroughs across a variety of domains, spanning image processing, to drug discovery and speech recognition – there are dozens of successful use cases, Perold of Element told Devex. “Ultimately, deep learning is a tool that must be applied thoughtfully,”

Technology should bridge gaps, not drive inequalities between those at the top and those at the bottom, said Perold.

“We believe that one of the best ways to do this is to meet people where they are – on the lowest common denominator of computing, the mobile device. Solving problems requires not just advanced technology, but a lot of thoughtfulness of context and application. It’s the responsibility of anyone delivering solutions to get to know the people, cultures, and delivery needs directly,” he said.

Read more Devex coverage on global health.