Artificial intelligence can help improve the lives of millions of patients. By Clare Milliken
Artificial intelligence is all around us. And if AI is not already in your doctor’s office, you can bet it’s coming soon — with the potential to create major improvements in our health care and well-being.
“AI is able to do things that we never thought possible,” says Northwestern professor and physician Abel Kho. “For example, with relatively little human input, AI can predict important clinical outcomes such as hospitalization or mortality risk. It can diagnose conditions. AI can cut the time it takes doctors to write notes or find important information.
“In short, AI is going to allow us to practice in ways that are simply better for patients,” says Kho, a professor of medicine and preventive medicine at Northwestern’s Feinberg School of Medicine and director of the Institute for Artificial Intelligence in Medicine (I.AIM).
When it comes to AI and medicine, “I generally talk about three areas,” says professor and cardiologist Sanjiv Shah. “One is diagnosis. There are many diseases, common and rare, that are misdiagnosed or underdiagnosed. AI can alert clinicians that they may be missing a diagnosis or that they have the wrong diagnosis.
“Automation is another huge growth area for AI in health care,” Shah says. “For example, on an echocardiogram [heart ultrasound], a cardiologist or ultrasound technician can make up to 100 measurements, which takes a long time — but the computer can do that automatically within seconds.
“The third area is classification: grouping or classifying patients who have heterogeneous diseases or conditions, which have varied root mechanisms but similar presentations. Even diseases we think we know — diabetes, hypertension, coronary artery disease — they each have varying underlying subtypes that require specific treatments to achieve optimal outcomes. By helping clinicians decipher heterogeneous medical conditions, AI can move health care toward more personalized precision medicine.”
Shah ’97, ’00 MD, the Neil J. Stone, MD, Professor at Feinberg, is now focused on using AI to classify cardiovascular disease patients. He’s just one of many medical researchers and clinicians across the University who are incorporating AI into their work. Others are using AI tools to uncover genetic underpinnings of disease, reduce prenatal stress and even allow patients with severe mobility impairments to gain greater independence and autonomy.
More than 6 million people in the U.S. have heart failure, which is among the leading causes of hospitalization in the country. As the population ages, an estimated 8 million people will experience heart failure by 2030.
The increasing prevalence of the condition is driven largely by a syndrome called heart failure with preserved ejection fraction (HFpEF), which is characterized by a stiffened heart muscle. Blood can’t pump into the heart efficiently, which results in pressure buildup in the lungs and elsewhere in the body, causing exercise intolerance and fatigue. The condition is associated with a high mortality and hospitalization rate. More than 50% of heart failure patients have HFpEF, and compared with other forms of heart failure, which respond well to current treatments, HFpEF has been much more difficult to treat.
“It’s been hard to find treatments because it is a heterogeneous syndrome with a variety of subtypes with different mechanisms and biological underpinnings,” says Shah, who in 2007 started the first clinic in the world dedicated to HFpEF study and treatment. “We need to move toward precision medicine to make more accurate diagnoses and tailor therapy to the specific subtypes.”
In 2014 Shah’s team used unsupervised machine learning (where AI finds patterns in unlabeled or ungrouped data) to sort through the information he had compiled on his HFpEF patients, including lab test results and data from physical exams, echocardiograms and electrocardiograms (EKGs).
The machine learning algorithm found three mutually exclusive, independent subtypes of HFpEF. “Nothing like that had been done in heart failure or in HFpEF or really in cardiovascular medicine,” Shah said on Northwestern Medicine’s Breakthroughs podcast in 2022. The use of machine learning “really took off after that, not just in cardiology but in a lot of other fields.”
Using machine learning, the researchers are now looking into the mechanisms behind — and potential treatments for — the various subtypes. In late 2023, Shah and his colleagues reported in The New England Journal of Medicine that the popular diabetes drug semaglutide reduced symptoms of heart failure in patients with one of the HFpEF subtypes that Shah’s team identified.
Shah is also using machine learning algorithms to find patterns in electronic health record data, echocardiograms and EKGs to help physicians diagnose HFpEF, its subtypes and related heart muscle diseases. “Once we’ve found a HFpEF subtype and have a treatment for it, we can use machine learning to train an algorithm to auto-diagnose or flag a patient with that specific subtype, so the clinician doesn’t miss the diagnosis and [can then treat] the patient appropriately,” he says.
“What we’ve done in HFpEF is applicable to so many medical conditions: diabetes, schizophrenia, depression, hypertension — you name it,” adds Shah, who is director of the Center for Deep Phenotyping and Precision Therapeutics within I.AIM at Feinberg and director of research at Northwestern Medicine’s Bluhm Cardiovascular Institute.
“There are a lot of skeptics of precision medicine — the right treatment for the right patient at the right time,” Shah says. “But I’m a believer. With the AI technologies we have today, we can identify subtypes within broad constructs of diseases, and that knowledge can be harnessed to create tailored treatments.”
Half a million people in the U.S. rely on power wheelchairs, and yet, due to the extent of their impairments, more than 15% of that population cannot use a conventional joystick to maneuver their chair. But AI can help wheelchair users safely navigate the world more independently.
Engineer and roboticist Brenna Argall is leading efforts to develop AI-enabled power wheelchairs for people with severe motor impairment, such as individuals living with multiple sclerosis or a spinal cord injury.
“The more severe a person’s motor impairment, the more difficult it can be for them to operate the machines that might enhance their quality of life,” says Argall, an associate professor of computer science and mechanical engineering at the McCormick School of Engineering. “The goal is autonomy for everyone.”
Power wheelchair navigation can be cumbersome and taxing for users with severe motor impairment. Furthermore, it can be difficult to adjust the wheelchair’s speed, putting users at risk of collisions.
“Despite decades of ‘smart robotic’ wheelchair research within academia and despite all the advances in driverless cars and other mobile robots, this technology did not translate to power wheelchairs until 2021,” Argall says. That’s when LUCI Mobility became one of the first commercial providers of a sensor-based safety system for power wheelchairs.
Argall, who is also an associate professor of physical medicine and rehabilitation at Feinberg, has partnered with LUCI to further expand its offerings for power wheelchair users. Her team developed and embedded an “active assistance paradigm” called REACT that can steer around obstacles autonomously.
With REACT, the artificial intelligence in the chair divides the surrounding area into eight zones and uses sensor data to calculate a safety score for each zone. If the user commands the chair to move to an area with an unsafe score, the chair will modify that command.
The chair’s sensors use millimeter- wave radar — the same technology used in vehicle crash detection and crash avoidance systems — as well as camera data to calculate depths and distances around the chair.
“What you get is a 2D grid of centimeter-resolution cells,” Argall says. The wheelchair’s AI operating system can “use that to reason about the world and build safe paths.”
The same sensor technology allows the chair’s intelligence system to construct digital maps, Argall says, allowing the user to “map out a known environment by driving the chair through the space” — the user’s home, for example. “Sensor information from the chair will build a map of that environment.” The user can also add location tags to the map — say, for the kitchen or bathroom — and then be driven autonomously to those locations.
And when a user is moving on that path to the kitchen or bathroom, the chair’s sensors account for dynamic obstacles in real time, such as a cat running across the hallway.
“A lot of this technology would be helpful for any wheelchair user,” Argall says, “just as driver-assist technologies on today’s cars are helpful even if you already know how to drive. Even for wheelchair users who don’t have severe motor impairments, this technology could still increase their access to the world.”
Prenatal stress can lead to perinatal depression, preterm birth and low birth weight, among other complications.
A Northwestern team led by Feinberg associate professor of preventive medicine Nabil Alshurafa is developing a machine learning algorithm that incorporates wearable devices and app-based surveys to predict a pregnant person’s next-day stress and offer tools to reduce and prevent stress.
“We have the tools to address stress in the moment,” says Maia Jacobs, an assistant professor of preventive medicine at Feinberg and of computer science at McCormick. “This new algorithm gives us a way to not only provide an intervention when a person is in the throes of stress but also to look for ways to reduce stress across the pregnancy.”
Jacobs and her team are working with Alshurafa to understand how best to embed the machine learning tool into the daily lives of pregnant people.
It’s just one of several projects that Jacobs is leading at the Northwestern Personalized and Adaptive Technology for Health Lab, where her team works directly with patients, as well as researchers, to explore how AI tools can be better utilized in health care settings.
“We are focused on understanding how people want to interact with these tools and how they can be designed to best support the end goals of patients, physicians and researchers,” says Jacobs, who is also the Lisa Wissner-Slivka and Benjamin Slivka Professor of Computer Science at McCormick.
Abel Kho has been working to improve data integration and data sharing, which, he says, are key to the successful implementation of AI in health care. After all, he says, AI is only as good as the data we give it.
Kho directs the Institute for Artificial Intelligence in Medicine (I.AIM) at Feinberg. Founded in 2020, I.AIM brings computational methods together with clinical expertise to advance health care and “be more efficient and scalable in how to make AI available for medical research,” says Kho.
Additionally, the institute’s Center for Medical Education in Digital Healthcare and Data Science integrates AI into Feinberg’s first- and second-year medical school curriculum and offers a “health data gymnasium,” an academic resource that allows students to interact with data. “We’re training future leaders in this space,” says Kho.
Photo Credit: Monika Wnuk ’14 MS, ’19 MS
When compared with neurotypical individuals, people with autism often show differences in “prosodic features” of speech, such as rhythm, intonation and pitch. Northwestern researchers are leveraging AI to better understand those differences and use that information to diagnose autism in people more accurately and at an earlier age.
“As early as the first delineation of autism in the 1940s, there were already reports of prosodic features being different in people with autism,” says Molly Losh, the Jo Ann G. and Peter F. Dolle Professor of Learning Disabilities at the School of Communication. “But 80 years later we still know very little about the nuances and origins of these prosodic differences. In particular, it’s quite hard to measure the differences scientifically, despite how apparent they may sound even to untrained ears.”
In collaboration with researchers at the Chinese University of Hong Kong, Losh’s team, including research assistant professor Joseph Lau, conducted a study comparing the speech patterns of English- and Cantonese-speaking young people, both with and without autism. The study focused on intonation, which is the way pitch is used in a sentence to convey different meanings, and rhythm, which refers to the articulation and intensity pattern of words and syllables.
Study participants were asked to tell a story from the wordless children’s picture book Frog, Where Are You?. Using unsupervised machine learning (where AI finds patterns in unlabeled or ungrouped data), the research team sought to determine whether the AI could differentiate participants with autism from participants without.
Across both languages, researchers found that differences in rhythm, but not intonation, were associated with an autism diagnosis. This suggests that rhythmic changes in autism may be more biologically or even genetically based.
The discovery that changes in rhythm are a hallmark of autism across languages “suggests those differences are more hardwired.” That’s meaningful for biological and genetic studies of autism, Losh says. “That’s a huge area of research for us.”
Because autism shows great variability across patients, Losh says the use of machine learning in autism research presents unique opportunities for a field that until now has relied heavily on in-depth interviews and researchers’ hand-coding of patient responses, both of which are time-intensive. Incorporating AI into this research could enable faster, more streamlined diagnoses of autism.
“Across prosodic speech features, some people with autism show sing-songy patterns while others might be monotone, and others might have completely different speech patterns,” Losh says. “Machine learning has the potential to pull out those fine-tuned differences and really help us understand them.”
That understanding, Losh says, will advance both biological study and clinical care and support for those with autism.
Artificial intelligence has allowed Alshurafa, Jacobs, Shah, Argall, Kho, Losh and many others at Northwestern to probe deeper into their work in the hopes of advancing both research and patient care. And with continued investment in and exploration of AI, the researchers are confident that we will see even greater benefits to come.
“AI will absolutely transform most aspects of care in very positive ways,” says Kho. “Diagnoses will be much more precise and aligned with therapies, and we’ll see tailored medications based on AI.”
“We’re trying to make a difference in the lives of our patients,” Shah says of Northwestern’s focus on bringing AI to real-world patient applications.
“It’s exciting to come up with AI algorithms, but if they don’t change how we practice medicine or serve the health of our patients, what good are they? We focus on building AI that will improve patient care.”
Clare Milliken is senior writer and producer in Northwestern’s Office of Global Marketing and Communications.
Check out these related stories...
Earth and planetary sciences graduate students Leah Salditch and Molly Gallahue spent a week in September hunting down earthquake anecdotes on California excursion. The memories they gathered will help inform state hazard maps of quake-vulnerable areas.
Northwestern psychologist Vijay Mittal says human behavior is made up of three primary components: emotion, cognition and motor activity. By examining motor behavior as both an early signal and a treatment tool, Mittal hopes to stop psychosis in its tracks.
As the world’s population rises and the effects of climate change threaten crop yields, food insecurity is an urgent matter. With the support of University Trustee Melih Keyman and Zeynep Keyman, Northwestern researchers are pursuing a novel approach to farming with the potential to make a substantial impact on the global food crisis.
Reader Responses
No one has commented on this page yet.
Submit a Response