Medical images are so detailed it can be hard to decipher them. Her program can spot what people can’t.
Medical images are massively important in diagnosing disease, but as they get more detailed it becomes harder and harder for a human being to interpret them. Shinjini Kundu created an artificial-intelligence system that can analyze them to find patterns undetectable to the naked eye. Her innovation could have a fundamental impact on the way we detect and treat diseases.
“If there are hidden changes and there is a way to detect these invisible patterns, then maybe we have a chance to diagnose diseases early, before symptoms develop,” she says.
There are already AI algorithms that teach themselves to spot patterns, but they’re not able to explain their reasoning. In medical diagnosis, this can be a limitation: without some knowledge of how and why a disease is developing, it’s impossible to address.
Kundu’s system allows humans to look through the eyes of the computer to discover otherwise imperceptible patterns that reveal the early disease process. She also trained the AI to pull out the disease markers from the images so that they can be seen on their own. That could help humans recognize them months or years before the onset of illness—so rather than just humans teaching AI, AI can teach us.
His method makes it possible to test 300 drugs at once.
For decades, the pharmaceutical industry’s approach to finding new cancer therapies has been to put tumor cells in a dish and test drug-delivering nanoparticles (particles between one and 100 nanometers in size), one by one by one, to find one that’s effective. Then researchers have to hope that those particles go where they’re needed when introduced into a living subject whose body might attack them or break them down.
“The problem is that, forever, people have been testing drug delivery vehicles the wrong way,” says James Dahlman, who runs a lab at Georgia Tech.
Dahlman has invented a radically different process that involves encoding each nanoparticle with a DNA sequence that he calls a bar code. Three hundred of those nanoparticles can be injected into a laboratory mouse, and when researchers remove the tumor, they can use gene-sequencing technology to determine how each of the bar codes did, all at once. The difference in volume is staggering. Dahlman says he tested about 30 particles during his entire PhD; in 2018 alone, his lab will hit 3,000. He hopes this technology could mean that a drug designed to treat a tumor in the lung, for example, could go straight to the problem area—rather than making the patient’s hair fall out.
Her filtration system could eliminate much of the energy used in industrial separation processes.
Shreya Dave thought her PhD research had no practical applications. It involved molecular filtration membranes made of graphene oxide—which is cheaper and less prone to degrading than the polymers and ceramics used today—but her method was too expensive for the water industry.
Then an article in Nature convinced her that the technique could save massive amounts of energy in the industrial processes used to separate chemicals for food, beverages, drugs, and fuel. These processes, it turns out, account for 12 percent of all US energy consumption.
Dave is now the CEO of Via Separations. The technology she and her team designed is meant to replace the current system for separating chemical compounds, which basically amounts to boiling. Dave believes that widespread adoption of Via’s filtration material could eliminate anywhere from 50 to 90 percent of the energy used in such industrial processes.
Her company is currently focusing on the food and beverage industry, but Dave thinks if she can prove that the technology is scalable and cost-effective in one industry, that will be the key to succeeding in others.
An earthquake led her to invent a blend of analog and digital technologies for use when networks are down.
When an 8.8 magnitude earthquake hit her native Chile in 2010, Barbarita Lara started tinkering. An engineering student at the time, she was struck by the challenges of communication in the quake’s aftermath: everyone she knew had become dependent on the internet and cell phones, but most networks were down. Along the Chilean coast, 156 people were killed in a tsunami triggered by the quake—in part because they didn’t receive a warning in time.
Eight years later, Lara has a product she thinks can help save lives in the next disaster. Her platform, known by its Spanish acronym SiE, allows smartphone users to receive messages from authorities via encrypted high-frequency audio: a blend of analog and digital technologies designed for use when internet and phone networks aren’t working. The SiE platform, which makes use of existing radio infrastructure, also enables smartphones to message each other using mesh, a radio-enabled wireless ad hoc network. Lara’s invention was inspired by Morse code, which her father, a cryptologist in the Chilean navy, introduced to her when she was a child. “Sometimes the best solution is very simple,” she says.
Emercom, the startup she founded to develop and market the platform, is now in discussions with Chilean disaster management authorities about the prospect of using SiE for future alerts. It’s also in talks with a leading telecom about pre-installing SiE on new cell phones.
—Jonathan W. Rosen
Hearing loss in humans has always been irreversible. His innovation may change that.
Will McLean believes he’s found a fix for a medical conundrum that many thought could never be solved: hearing loss in humans.
McLean’s research focuses on the cochlea, the spiral-shaped cavity within the inner ear that’s responsible for hearing. At birth, the average human cochlea contains 15,000 hair cells, which detect sound waves and transfer them to the brain. Over time, many of these cells are killed by exposure to loud noise and toxic medications. In mammals, unlike birds, reptiles, and amphibians, they don’t naturally grow back. “The inner ear is one of the least regenerative parts of the body,” McLean says. “That’s why hearing loss is permanent.”
McLean, who holds a PhD from MIT in health science and technology, has spent the last decade trying to change that. His early work showed that the inner ear contains distinct progenitor cells—similar to stem cells but more specific in their capabilities—and that some have the potential to become hair cells, though they cannot divide or differentiate on their own to repair damaged tissue. To resolve this, he and colleagues used insights from regenerative tissues, such as those in the intestine. They exposed damaged cochleas from mice to a combination of drugs that can trigger regeneration in these other organs. Surprisingly, their technique not only caused the progenitor cells to proliferate but also induced them to generate new hair cells—the key to restoring hearing.
On the strength of this discovery, McLean and colleagues established Frequency Therapeutics, a startup working to commercialize what he describes as an entirely new mode of medicine. Frequency’s technique, known as progenitor cell activation, uses a combination of compounds that essentially unlock the body’s ability to heal itself. To date, Frequency has filed 19 patent applications and developed an injectable in-ear therapeutic to combat hearing loss. The treatment has successfully passed human safety trials.
—Jonathan W. Rosen
His computer chips mimic the workings of the human brain.
Manan Suri has built key elements of computer chips that mimic the learning ability and energy efficiency of the brain. And he did it by harnessing a quirk of next-generation memory technology.
That technology is known as emerging non-volatile memory (eNVM). Because of peculiarities in their nanoscale physics, eNVM devices often behave in random ways, which in computers is usually a flaw. But Suri realized that this irregularity could help researchers build so-called neuromorphic chips, which emulate the neurons and synapses in our brains.
While transistors store information as 1s and 0s, the biological synapses that store information in the brain can take multiple states. That means building computers that behave like the brain traditionally required complicated artificial synapses that can also take multiple states.
Suri recognized that he could harness the inherent variability of eNVMs to build large-scale neuromorphic systems capable of doing supervised and unsupervised learning. He’s exploited that irregular behavior for cybersecurity and advanced sensing applications. Earlier this year he founded a startup, Cyran AI Solutions, to build neuromorphic and cybersecurity hardware based on his eNVM research.
Making off-the-shelf electronics stretchable.
Stretchy electronics that can conform to the body no longer have to compromise between electrical and mechanical performance, thanks to some smart engineering by Sheng Xu.
Marrying rigid electronic components with elastic materials is tricky. The mismatch in their mechanical properties generates huge strains, causing them to separate when deformed. That’s why most previous research in flexible electronics focused on building new components that are soft and flexible. But Xu didn’t see the sense in discarding decades of progress in the electronics industry. “Why not use something that already matured decades ago?” he says. His strategy made it possible to integrate off-the-shelf components into elastic materials to create highly stretchable electronics as capable as their rigid counterparts.
Xu opted to bond only tiny sections of the components to the elastic material and then support them in a fluid-filled capsule. These are joined together with wires configured into long wavy lines that unravel in an ordered way when stretched. He’s used the approach to build a lithium-ion battery that stretches by up to 300 percent and a hospital-quality health monitor that conforms to the body as it moves. The latter has been developed into a wearable physiological sensor called BioStamp by a startup called MC10.
Her innovations could make better, cheaper alternatives to silicon solar cells.
the solar energy industry has lacked a low-cost, high-performance alternative to silicon for a long time. In recent years, a family of hybrid materials called perovskites has gained attention because they can achieve high power output more cheaply than silicon. But making them work in practice has proved difficult. Early prototypes of perovskite-based solar cells weren’t as efficient as conventional silicon cells at converting the energy in sunlight into electricity.
Huanping Zhou developed a series of chemical processes that made perovskite-based solar cells more efficient and cheaper to produce. If they can be mass-produced, her innovation will make solar power much cheaper.
Growing up in the countryside of China, Zhou did not have electricity at home. She and her siblings did their homework by the light of a kerosene lamp. Her childhood experience motivated her to devote herself to solar technology.
The cell Zhou developed converts more than 20 percent of the energy in sunlight, about the same rate as existing silicon panels. Although some other perovskite cells are more efficient, Zhou’s invention is important because it makes the manufacturing process easier and cheaper. The cells can be produced at temperatures below 302 °F (150 °C) by spraying or printing a perovskite-based liquid solution onto a substrate such as glass. The process for some other types of perovskite cells requires temperatures around 932 °F.
Perovskite-based solar cells tend to degrade faster than silicon cells, so Zhou is also working on improving their durability.