Millions of Americans know they're sick, but no doctor has been able to diagnose them. Many spend years — sometimes decades — seeing specialists and having invasive tests, only to be told that their results are "within the normal range." Some are referred to psychotherapists and psychiatrists. Others just give up.
Will artificial intelligence be a game-changer for these patients?
It's possible, according to UTSA biomedical engineer Amina Qutub, PhD. Qutub is an associate professor, leads a lab at UTSA, and is an assistant director for the UTSA Matrix Consortium, a group of AI researchers, applied engineers and scientists who are focused on finding ways to use AI to improve human well- being.
Qutub said AI applications like ChatGPT are already helping people help their doctors figure out what's going on with their health.
"ChatGPT and other algorithms are generating predictions of people's medical conditions," Qutub said, "and as you add information to ChatGPT, as you provide your data, it's trying to find patterns that it can link up to prior cases."
When AI makes these connections, it can offer a user suggestions about possible diagnoses.
Jacob Rosati thought he'd give it a try.
Rosati is a composer, sound designer, and a producer for TPR's "Petrie Dish" podcast. He also lives with an undiagnosed condition.
"About 13 years ago, I woke up one day feeling sort of fuzzy, as if I'd been hungover, but I hadn't drank or anything," Rosati typed. "Everything felt far away from me in a strange way, like looking at the world through a window. I could clearly see, but it just felt wrong."
Rosati explained that not long after that first day, these symptoms became his daily reality. For 13 years, no doctor has been able to offer him a diagnosis or any treatment that alleviates the symptoms.
"I had EKGs, EEGs, MRIs," he said. "I went to several eye doctors. I was prescribed antidepressants. I was prescribed migraine medication. I did an elimination diet. I did a silent meditation retreat. I tried acupuncture, a lot of therapy. I went completely sober—not that I was using a lot of substances before. I tried psychedelics—that didn't do anything—and many more things I can't really remember."
ChatGPT is what’s called a large language model — a type of artificial intelligence that learns how language works by ingesting a huge amount of data, including an incalculable number of books and articles from throughout history and all over the world.
When someone like Rosati asks a question, ChatGPT has all of these resources at its disposal, Qutub said.
"It will look for patterns in history, patterns from published data, or patterns in the data, and they'll flag that. They'll say these are really interesting, or we think this is what you might be suffering from," Qutub explained.
Rosati received several suggestions in response to his question, including something he'd long thought might be a part of his diagnosis.
"Derealization disorder," Rosati said. "This condition involves persistent feelings of detachment or being outside oneself; depersonalization and experiencing the world as unreal or distant."
"Derealization, that's how I feel," he said.
Rosati said he first stumbled upon the word "derealization" about four years ago, after nearly a decade of searching for answers.
"I don't know why it took me so long, but I think that's pretty cool, actually, that I didn't mention derealization in my initial prompt [to ChatGPT], and it brought me to it," he said.
No doctor ever mentioned derealization as a possible diagnosis for Rosati, and Qutub says it's common for people with symptoms that don't trigger quick answers using blood tests or other diagnostic tools to feel like their doctors have dismissed their concerns.
"One of the critiques and frustrations from those who suffer from many different chronic disorders where they don't have a diagnosis and they don't have a treatment is that they're often discarded or feel that their voices are not heard from the medical community," Qutub said.
Artificial intelligence could find answers that would force the medical community to take these patients seriously, Qutub added, proving their experience is not all in their heads.
"These are real symptoms. These are quantifiable symptoms. These are quantifiable characteristics that people are suffering from," Qutub said.
Qutub is currently working with artificial intelligence to better understand chronic Lyme disease. Other disorders that are diagnosed using symptoms and patient history, not a blood or genetic test, include long COVID, Ehlers-Danlos syndrome and other hypermobility spectrum disorders, and fibromyalgia. Artificial intelligence may help researchers develop tests and better treatments for diseases like these.
To learn more about AI's potential role in solving medical mysteries, listen to this episode of Petrie Dish featuring UTSA associate professor and researcher Amina Qutub.
