The Rise of AI Tech expert: 'It's the age of AI right now in health care'

Oct. 28—JOHNSTOWN, Pa. — During a routine visit with the medical oncologist, a cancer patient brings up a new symptom that could be a side effect of the cancer treatment.

The conversation is being recorded by an artificial intelligence model using ChatGPT, which also hears the doctor say he will recommend a new medication to help with the side effect.

The software then prints out a summary of the conversation, along with the patient's latest lab work, radiology reports, vital signs and clinical information from the visit in a form identifying billing codes that is understandable to insurance companies.

Across the hospital, in the radiology unit, a patient from the emergency department is getting a computed tomography scan of the brain with technology that includes artificial intelligence software that flags features showing the patient has recently had a stroke.

In another area, an AI program is automatically "scrubbing" patients' electronic medical records to flag patients at risk of developing a bloodstream infection known as sepsis, which killed more than 200,000 people in the United States in 2019, according to the Centers for Disease Control and Prevention.

All of these examples of artificial intelligence are already improving patient care at hospitals across the country, and there is plenty more on the horizon, experts say.

"It's really going to revolutionize medical care and benefit every one of us," Dr. Dajiang Liu said from Penn State College of Medicine.

As professor and vice chairman for research in the school's Department of Public Health Sciences, Liu leads medical AI research for the university.

Liu said the realm of medicine is a natural fit for artificial intelligence because AI and medicine are both built on data.

"Nowadays, biotechnology in medicine has moved to a stage where we can generate a huge amount of data without too much effort," Liu said. "Now the challenge is how to make sense out of the data to help improve health care and biological research."

A group of leaders in the field described the historic use of data in medicine in the introduction to their paper, "Meeting the Moment: Addressing Barriers and Facilitating Clinical Adoption of Artificial Intelligence in Medical Diagnosis," published last year by the National Academy of Medicine.

"Clinical diagnosis is essentially a data curation and analysis activity through which clinicians seek to gather and synthesize enough pieces of information about a patient to determine their condition," the report reads. "The art and science of clinical diagnosis dates to ancient times, with the earliest diagnostic practices relying primarily on clinical observations of a patient's state."

With AI and machine learning, software systems collect and analyze data from multiple sources that affect an individual patient to provide a backup for the physicians who are treating them.

"It's a tool that helps them provide better care," UPMC executive Dr. Salim Saiyed said. "It doesn't take over. It enhances our ability to detect, identify and intervene. It's all about how you implement and how you use it."

Saiyed is vice president and chief medical information officer at UPMC in central Pennsylvania. He leads the implementation of clinical technology, optimization, informatics, governance, analytics and digital health.

"It's a super-exciting time to be in health care," he said. "AI gives us the ability to do a lot of mundane tasks, as well as be able to predict and take better care of patients with personalized treatment. It's the age of AI right now in health care. It's really revolutionizing how we provide care to patients."

By using ChatGPT technology to help take and transcribe notes, doctors can spend more time with patients and improve care, Saiyed said.

"It's not a straight-up transcription," he said, noting that the software merges information from several sources.

"The AI is smart enough to put all that together," he said. "It knows what part of those conversations are important and where things go. It tremendously changes how we chart. It enhances that physician-patient encounter."

Many of the emerging models are being built into radiology systems to analyze medical imaging, including X-rays, mammograms, ultrasounds, CT scans and magnetic resonance imaging.

"Radiology is a leading medical subspecialty in terms of research development and translation of AI in the medical domain," Dr. Shandong Wu said from the University of Pittsburgh School of Medicine. "These new techniques have been shown that they are able to read or interpret those images with a comparable or even better performance when compared to the performance of a radiologist."

Wu is director of the Intelligent Computing for Clinical Imaging lab and technical director for AI innovations in radiology at the University of Pittsburgh School of Medicine.

At Conemaugh Memorial Medical Center in Johnstown, radiologists are using AI systems to help analyze 3D mammograms and breast MRIs to identify features for further evaluation, Radiology Director Clifford Dull said.

"There is a lot of information in a 3D mammogram," he said. "Computer-aided diagnosis draws a circle on the image to help show what you want to look at."

Radiologists then review the image, with additional attention to marked areas, to determine a final diagnosis, a recommendation for a biopsy or for more imaging tests.

Also at Conemaugh, the software application RapidAI checks CT brain scans for ischemia, or inadequate blood supply to a part of the body. It then creates an image, marking the affected area and alerting the radiology team with a stroke alert.

"All of this is done without human interaction," Dull said. "It's pretty nifty."

Another system employed in Conemaugh's Healthy Persons program looks at all the radiology reports for each patient's X-rays, MRIs, ultrasounds and CTs. It then flags certain words for review by the radiology department with suggestions for follow-up care.

It might be a lung screening that identifies a nodule in one lobe.

"It's going to kick that out and it will go into a queue to put that patient in a surveillance program," Dull said, explaining the patient would be called back for another screening in six months or a year to see if there is any change.

The system also monitors breast, thyroid and pancreas images, as well as patients with small aortic aneurysms.

"It's like a stop-gap program that helps keep the patient from falling through the cracks," Dull said.

Conemaugh Health System is cautiously examining new AI systems as they are introduced, he said.

"AI is still in its infancy," Dull said. "It is a very big endeavor and changes the work flow. It requires IT infrastructure changes."

The experts recognize that there are ongoing challenges for AI in medicine. A new study led by Stanford School of Medicine researchers cautions that popular chatbots are perpetuating racist, debunked medical ideas, prompting concerns that the tools could worsen health disparities for Black patients, the Associated Press reported earlier this week.

Researchers found that chatbots such as ChatGPT and Google's Bard responded to the researchers' questions with a range of misconceptions and falsehoods about Black patients, sometimes including fabricated, race-based equations.

Experts worry these systems could cause real-world harms and amplify forms of medical racism that have persisted for generations as more physicians use chatbots for help with daily tasks such as emailing patients or appealing to health insurers.

In some cases, they appeared to reinforce long-held false beliefs about biological differences between Black and white people that experts have spent years trying to eradicate from medical institutions.

Those beliefs are known to have caused medical providers to rate Black patients' pain lower, misdiagnose health concerns and recommend less relief.

"There are very real-world consequences to getting this wrong that can impact health disparities," Stanford University's Dr. Roxana Daneshjou, an assistant professor of biomedical data science and dermatology and faculty adviser for the paper, told the AP. "We are trying to have those tropes removed from medicine, so the regurgitation of that is deeply concerning."

Wu said researchers are aware of some of the shortcomings and are working to be sure the models are inclusive.

"We have a responsibility to make these AI models unbiased," Wu said. "I think people will accept it, but there are also limitations. There may be under-represented groups in the data. How do you make sure your model is ethical and widely uses all races? These are the issues in front of us."

One caution, he said, is that software does not consider everything that could affect a condition.

"This has to be carefully interpreted," Wu said. "It focuses on a certain kind of data. Some models may not yet fully interpret it all, but eventually tools will be developed. This is not just for fun."

Even as they address those issues, AI proponents have to make sure doctors and hospitals are comfortable using the new technology.

"As with any new technology, there is always apprehension early on," Saiyed said, noting that health care professionals are used to embracing new technologies and treatments if they are shown to be effective. He points to the growth of robotic surgery over the past couple of decades.

Part of the hesitation, Wu said, is what those working in AI call the "black box" problem, in which a system's inner workings are not clear to users.

"They don't know how it works. They say, 'How do they do this? I don't understand,' " he said.

Researchers get it, he said, and they are already working on ways to help people understand the software's methods.

"If people know how this works, how the model made this decision, that will increase the confidence," Wu said. "A lot of researchers are trying to make the models more transparent and explainable."