F E A T U R E
F E A T U R E
Pushing the Boundaries With AI
By Andrew Faught • Illustration by Ard Su
Three faculty members are using AI to tackle skin diseases, improve understanding of how scent impacts people with autism, and discover solutions to treat neurological and psychiatric conditions.
From his lab in Theobald Science Center on the Long Island campus, Claude Gagna, Ph.D., could shape up to be your favorite workout partner.
That’s because the New York Tech professor of biological and chemical sciences is using artificial intelligence (AI) to battle the bane of exercise enthusiasts everywhere: athlete’s foot and other fungal skin infections.
His work could help create new antifungal medications—personally tailored medications, at that—as some traditional treatments have lost their effectiveness over the years against the increasingly drug-resistant skin diseases, Gagna says.
“AI swiftly and correctly identifies fungal organisms that invade skin,” he adds. “It also helps identify the type and extent of the skin pathology. And then we can use this data to eventually personalize medicines based on a patient’s genetics.”
Gagna is one of many New York Tech researchers who are using AI to push the boundaries of medical understanding and innovation. They’re doing so by collecting data (DNA in Gagna’s case) and then processing the information through algorithms—instructions that a computer program uses to analyze data, identify patterns, and make predictions.
Several other New York Tech researchers are using AI and its subset, machine learning, which helps computers improve from experience without being explicitly programmed, to study autism and mental health disorders.
Scratching an Itch
As data surges by the second, AI plays an integral role, addressing what ordinarily amounts to data overload for the human brain. Every day, humans and machines across the globe create 2.5 quintillion bytes of new data, according to IBM—enough to fill 20 billion file cabinets. Making sense of all that information is the work of AI. For his research, Gagna gets his data from medical records and tissue biopsies.
The research project focuses on dermatophytes, the nettlesome fungi that cause athlete’s foot, jock itch, and ringworm. The infections thrive in humid conditions, such as stinky athletic socks, creating crusty blisters and ooze. These infections could be deadly in immunocompromised patients.
“I’m teaching AI to recognize different shapes of dermatophytes at different life-cycle stages, and this is going to be a continuing process,” Gagna says. “The more I educate AI with different examples, the smarter it will get.”
Dermatophytes live on the skin, hair, and nails, mostly staying out of trouble. But when they interact with sweat and enter abrasions on the body, fungal outbreaks can occur.
As the next stage of his research, AI will help Gagna divine even more secrets of “exotic” DNA, which have unusual structures beyond the more commonly understood double helix. These nonconventional DNA structures, such as triple-stranded, four-stranded, cruciform, and hairpin DNA, regulate gene expression. He’s aided in his efforts by New York Tech B.S./D.O. (life sciences/osteopathic medicine) and biology students, who obtain microscopic images of dermatophytes, which Gagna then trains AI to identify and differentiate from normal cells.
“What better way to neutralize fungal organisms than to attack these specific types of DNA?” Gagna says. “This is a paradigm shift from the current focus of targeting the organism’s cell membrane.”
“You educate AI from day one, give it more and more information, and then it becomes more and more knowledgeable and oriented toward doing a perfect job.” –Claude Gagna, Ph.D.
These “higher order” structures could yield additional insights not just for skin conditions but also for cancers and other life-threatening diseases.
“Most scientists think of DNA as a stagnant one-dimensional molecule, the one that Watson and Crick discovered,” Gagna says. “Researchers need to go beyond that mode of thinking and consider the possibility of many different exotic DNAs that regulate gene expression in all types of cells. This approach can then be used for the treatment of cancer and other diseases.”
AI is in its infancy, and Gagna has had to familiarize himself with bioinformatics software created to recognize dermatophytes in stained human skin tissue that is typically analyzed by a human under a microscope. He then gathers mountains of data that will serve as a baseline for future machine learning. This will help dermatopathologists to properly diagnose the extent of the skin pathology.
“It’s basically like having a newborn,” Gagna explains. “You educate AI from day one, give it more and more information, and then it becomes more and more knowledgeable, and oriented toward doing a perfect job.”
Claude Gagna
On the Scent
For Gonzalo Otazu, Ph.D., AI is helping to unravel the mysteries of brain function. He’s using a machine-learning approach to understand olfactory—or smell—disorders among people affected by autism spectrum disorder. Smell, like other sensory information, is an important social cue for individuals with the condition.
Those with autism are considered to have both heightened and diminished olfactory function. While related behaviors have been documented in other sensory modalities, less is known about the neural processes leading to such behaviors.
AI has posed challenges in the early stages of research. Otazu, an assistant professor of biomedical sciences at the College of Osteopathic Medicine (NYITCOM), continues to gather research to feed into the computer. But algorithms designed by humans, which were designed to mimic certain aspects of the human brain, still aren’t as nimble as the real thing.
For now, AI struggles with the “novelty” of the stimulus. If a stimulus was part of the vast amounts of data that the AI was trained with, there is no problem. Problems arise if the stimulus is novel, i.e., the stimulus was not part of the data that an AI trained with.
“The questions that we have are more computational,” he says. “With current AI, if you have data for machines to train with, they’re pretty good at it. And while the brain can deal with novelty, machine learning doesn’t deal as well with it.”
“I think this is going to revolutionize how we think about brain function while yielding advancements in machine learning.” –Gonzalo Otazu, Ph.D.
Otazu is studying the brain circuitry of mice, whose livelihoods depend on scent more than any other sensory function. In his research, he used neurotypical mice and mice with a mutation in a gene linked to autism.
While both groups of mice were able to identify a target scent, those with the gene mutations struggled to identify scents when they were released against novel background odors. The inability to filter out smells suggests a brain-processing error, Otazu reported in 2023 in the journal Nature Communications.
Students helped Otazu build machines that were able to generate complex odor patterns.
His research could ultimately explain the brain circuitry of people with autism, which could lead to efforts to improve the quality of life of autism spectrum disorder. Scents can be a source of distress for people with the condition. Otazu says his research signals the need for further studies.
“I’m looking at the most advanced computer that we know, which is the brain, and trying to figure out what the brain is doing,” he says. “Right now, we have some ideas—big ideas—about connections in the brain. The big push now is to create even more comprehensive and detailed maps about those connections.”
Otazu says his work is laying the groundwork for bigger advancements to come. “I think this is going to revolutionize how we think about brain function while yielding advancements in machine learning.”
While AI was first described in medical research in the 1950s, technological limitations meant it was still a half century from becoming reality. Limitations were overcome in the early 2000s with the advent of machine learning, also known as deep learning, according to the National Institutes of Health.
AI is proving to be a boon to researchers, but it’s also revealing the brain to be a concurrently potent force.
“Current machine learning is data hungry,” Otazu says. “You need to give it tons and tons of data, whereas the brain can create useful actions with much more limited data. That’s one of the big questions in the field: How does the brain use limited data so well, and why do we need so much data for machine learning?”
Gonzalo Otazu
In Search for a Sound Mind
Algorithms are also helping to better identify mental health conditions and the sources of life-threatening seizures. Research by Maryam Ravan, Ph.D., research assistant professor of electrical and computer engineering, could help physicians better pinpoint where seizures originate, for example.
Separately, her work could also be used to help create noninvasive solutions that would allow clinicians to treat neurological and psychiatric conditions more effectively.
Developments are happening at a steady pace.
While epileptic seizures are known to be caused by brain injury and genetics, the cause is unknown in 70 percent of epilepsy patients, according to the American Association of Neurological Surgeons. Further, the source of the seizures is often difficult to localize in the brain.
Seizures are typically caused by brain wave imbalances, which can be detected by electroencephalograph (EEG), whose sensors detect electrical activity on the scalp. But EEG readings have a high margin of error. Another diagnostic tool, magnetic resonance imaging (MRI), is expensive, and some patients—such as those with electronic devices, including pacemakers and defibrillators—are cautioned against the technology.
In addition to those technologies, doctors use brain source localization (BSL) techniques to estimate the source of a seizure. One of the most widely used BSL techniques is a mathematical approach called exact low-resolution brain electromagnetic tomography (eLORETA). The technique improves the conductivity of tissues, offsets geometry uncertainties (when MRIs are not available), and aids in EEG electrode positioning.
To address these sources of uncertainty and improve the accuracy of source localization, in a collaborative study with researchers from Stanford University and Staffordshire University in the United Kingdom, Ravan introduced a robust version of eLORETA named ReLORETA. This new algorithm can adaptively deal with different uncertainties, regardless of their nature.
It was described in the journal IEEE Transactions on Biomedical Engineering, published in September 2022.
“ReLORETA provides precise localization of brain activity, which can lead to more accurate prediction by AI models,” she adds.
“Researchers in biomedical engineering are experiencing extraordinary excitement as AI rapidly transforms the field.” –Maryam Ravan, Ph.D.
Ravan has also developed deep-learning and machine-learning algorithms that analyze brain waves and categorize patterns as biomarkers for schizophrenia, bipolar disorder, and major depressive disorder.
Currently, neither condition has any known biomarkers, and it’s not unusual for doctors to prescribe the wrong combination of treatments because the conditions present similar symptoms.
“In the coming years, advances in AI applications in neural engineering, such as multimodal data integration and wearable technology, are expected to revolutionize the understanding and treatment of mental and neurological disorders,”’ Ravan says, noting that New York Tech master’s and Ph.D. students are involved in the work. “We’re making these algorithms more robust,” Ravan says. “Researchers in biomedical engineering are experiencing extraordinary excitement as AI rapidly transforms the field.”
Maryam Ravan PHOTO: ANDRE KOPINSKI
New York Institute of Technology Magazine © 2024 All Rights Reserved