Innovation in Health Care | Minnesota Monthly

The year is 2045. You’ve been feeling slightly off, so you make an appointment to see the doctor. You step into a gleaming medical facility, where holographic guides greet you by name and instantly access your medical history. After a painless blood draw, diagnostic machines powered by artificial intelligence (AI) analyze your samples in real time, providing immediate, accurate results and suggesting the most promising course of treatment.

During surgery, AI-guided robotic assistants perform with unparalleled precision while nanobots monitor your recovery and dynamically adjust your personalized care plan. Back at home, wearable devices and smart home technology keep you continuously connected to your care team as you heal. This vision of AI-saturated health care may seem fanciful, but it’s rooted in an established history of machine learning being used in medicine, as well as quickly evolving technologies that stand to revolutionize—or at least greatly augment—care.

“We’ve had AI systems in our health care environments for decades, kind of living under the radar,” says Dr. Ryan Jelinek, medical director for telehealth and patient access at Hennepin Healthcare. “Now they’re becoming more publicly noted because of breakthroughs in generative AI, specifically large language models like ChatGPT and Google Gemini. With that, a lot of novel applications are coming across our desks.”

Unlike traditional AI, which analyzes data to make predictions or classifications, generative AI can create novel outputs such as text, images, and unique insights. This is opening up new horizons of possibility for health care, from developing new drugs and personalizing treatment plans to interpreting medical imaging and streamlining administrative tasks.

New applications of AI are becoming more visible and pervasive at health systems across the state. These advances are not only pushing the boundaries of what’s possible in health care, but also making AI an integral part of the patient experience. In Minnesota, leading health systems such as Mayo Clinic, Allina Health, Hennepin Healthcare, and M Health Fairview are at the forefront of integrating AI into their medical practices. Read more about how Dr. Mark Lobanoff, of OVO LASIK + LENS, is shaping the future of ophthalmology with cutting-edge AI solutions.

Earlier Detection and Diagnosis

During the early months of the COVID-19 pandemic, when diagnostic tests and emergency room beds were scarce, a team of researchers at the University of Minnesota worked with M Health Fairview to develop an AI algorithm to evaluate chest X-rays and diagnose possible cases of COVID.

Learning from thousands of X-rays, the model developed the ability to identify COVID-19 in seconds. It also shared a risk score with providers. “The scoring system enabled us to see who was stable enough to go home with supportive care, and who needed to be admitted,” says Dr. Sameer Badlani, who leads enterprise strategy, digital, and experience functions for Fairview Health Services. The algorithm proved to be so useful, it was deployed across all 12 M Health Fairview hospitals and made available at no cost to other hospital systems. The risk scoring system was also expanded to other diagnoses, helping providers prioritize emergency care for those most in need.

Assisting in diagnosis is one of the areas in which AI shines. “The goal of medical care, in an ideal state, is trying to prevent disease from happening in the first place, or identifying it early in its course,” says Dr. Suraj Kapa, a digital health expert and cardiac electrophysiologist at Mayo Clinic. “That can be difficult for highly complex diseases that traditionally require a lot of specialized testing that entail additional costs and effort.” Some of those complex conditions include heart conditions (such as atrial fibrillation and hypertrophic cardiomyopathy) that can be difficult to identify on a standard electrocardiogram (EKG) and often require advanced—and expensive—testing to diagnose. In the absence of specific symptoms suggesting the need for more testing, people with these conditions can slip through the cracks of EKG screening.

But that may be starting to change, thanks to AI. “There are very subtle features in the EKG that can point to these more complex diseases. By training computers on EKGs of patients who do and do not have these conditions, the models learn to pretty reliably determine if someone might have this disease process that might require more intensive evaluation,” Kapa says.

Recently, a patient visited a clinic with shortness of breath. An initial ultrasound looked normal, and an EKG was interpreted by a clinician as normal as well. The patient was referred to Mayo Clinic, where another EKG run through the AI model noted a risk of amyloidosis (abnormal protein deposits accumulating in the heart). “The test showed a high probability of amyloidosis even before we were able to get additional testing, so we could more quickly decide on treatment,” Kapa says. Because basic EKG functions have become widely available via tools like smartwatches and Fitbits, these models could even theoretically be loaded on a smartphone and flag concerning signals without a clinic visit, as long as they were configured to the data source.

Nowhere is early detection more critical than in cancer care. Allina Health has implemented a broad-based imaging platform called Ferrum AI, aimed at using AI to supplement standard human reading of radiology images in order to identify concerns that human reviewers may have missed. Different modules can be used to examine breast mammography, prostate MRIs, liver CT scans, and chest X-rays. It has proven to be particularly useful in looking at lung CT scans, says Dr. Badrinath Konety, president of Allina Health Cancer Institute. “We look at scans that have been obtained incidentally, and we find sometimes that there are lung nodules that the radiologists may have missed or thought were not significant.”

In 99.7% of cases, the AI doesn’t identify anything new, or the radiologists confirm that the findings are insignificant. “But that less than half a percent of scans that do identify something new lead us to reevaluate and make sure we’re not missing anything,” Konety says. Across thousands of scans, even half a percent could equate to hundreds of missed lung cancers being identified. “That’s remarkable, because now you’re finding patients at an early stage, when the disease is much more curable and the outcome is much better. Adding AI to these screening programs is only going to enhance earlier detection,” Konety says. And as the models continue interacting with clinicians and learn more about patient outcomes, they’ll keep refining their judgment on what to flag.

Allina is also conducting a clinical trial in partnership with biotechnology company Astro Biosciences, using a combination of AI and holographic imaging to identify circulating tumor cells in early-stage breast cancer patients. The approach entails using a laser beam to create a 3D holographic image of cells from a blood sample. An AI model then compares these images to a vast database of cancer and normal cells to identify potential cancer cells.

“We’re trying to see if we can identify patients who have various types of cancers much earlier than even imaging would be able to detect them,” Konety says. While the study is still underway, early data appears promising.

Unlike other approaches that look for pieces of DNA emitted by cancer cells, this trial is looking for whole cells, which provide more comprehensive information. “You can grow the cells in the lab and test them with different types of chemotherapy to see what drugs may or may not work,” Konety explains. This approach may also discern between more and less aggressive types of cancer cells, which could help identify which patients need more intensive treatment to prevent recurrence. Ultimately, these advances may translate to better treatment outcomes and saved lives.

Enhanced Communication and Convenience

When their daughter Faith was born four months prematurely, Nick and Shacreya Lee of Isanti, Minnesota, had to organize their days around visits to the neonatal intensive care unit (NICU) at M Health Fairview Masonic Children’s Hospital. This was made considerably easier by a new AI-enhanced tool called Q-rounds.

Co-developed by Dr. Michael Pitt, a pediatric hospitalist at M Health Fairview Masonic Children’s Hospital, Q-rounds addresses the common issue of families missing critical rounding schedules due to lack of information. With a combination of AI and the provider’s own decision making, patients are marked high-priority or ready for discharge in a rounding schedule that can be shared with families, nurses, interpreters, and anyone else who needs to be present when the doctor arrives. The app sends real-time notifications, updating people on where they are in the queue.

Family members can even join visits remotely if they’re not able to be there in person. So, families like the Lees can plan their time more effectively between hospital visits, work, and home responsibilities. The tool, piloted successfully in the NICU, nearly tripled family presence during rounds. “It’s been a phenomenal solution for us in the NICU, which is one of the most crucial places you want families, nurses, and doctors to be [able to connect],” says Badlani.

Since its launch in 2023, more than 1,000 families have used Q-rounds to participate in rounds virtually. The increased presence of nurses during rounds has led to a nearly 40% reduction in harmful errors.

Improving the Patient-Provider Connection

We’ve all had a doctor visit in which the provider seems more interested in their computer screen than in us. Beginning in the 2000s, it became commonplace for providers to have to type notes into the electronic health record throughout patient visits. Many of them dislike this as much as patients do.

Provider burnout has become a critical issue, driven in part by these kinds of administrative tasks and the way they detract from the patient-provider connection. Health systems are exploring how AI can help reduce the burden of documentation that’s weighing on physicians, helping to fuel a shortage of primary care doctors.

Hennepin Healthcare has deployed an AI tool that creates clinical notes on patient visits, allowing providers to focus more on meaningful, face-to-face communication. Embedded in the provider’s smartphone, the tool records patient-provider interactions, extracts the most relevant information, then generates a clinical note—something providers previously had to do manually.

Currently used in Hennepin Healthcare clinics, the tool is expected to expand into emergency and inpatient settings later this year. “Around 70% of providers are using it routinely,” says Dr. Ryan Jelinek. “Some have found it life-changing in terms of the amount of time they’re having to spend on the electronic health record after work or during non-clinical hours. They’re reporting a better ability to just focus on their patients during the short, precious time they have together, so it’s a more quality experience for both of them.”

Patients are enjoying it, too, reporting that the physicians seem less distracted by having to type into a computer while they talk. As with other health care tools, including the electronic health record, the clinic notes tool is operated by an outside organization, giving it access to potentially sensitive patient information. This means that ensuring robust protocols are in place for patient privacy and data security will remain paramount as the technology evolves. “This is a nascent technology that’s going to change a lot and keep getting better,” Jelinek says.

Cautions and Risks

AI-driven diagnostic advances and enhanced connection and convenience stand to benefit both patients and providers. But there are risks to consider as well.

Cost is a big one. “These technologies aren’t free. They’re very expensive and they require a ton of energy to use,” Jelinek says. As in all industries, rising health care costs find their way back to the consumer. “We have to be cognizant of the associated costs and what the return on investment is.”

There are also built-in risks associated with sharing patient data with third parties, and the trustworthiness of AI tools has room for improvement. “One of the risks of generative AI is what’s known as hallucination. Because it can generate and transform information, you may get inaccurate information being presented with a level of sophistication and conviction that would lead you to believe that it’s accurate,” Badlani says. (AI tools such as ChatGPT sometimes cite scientific articles that sound real, using real author and journal names, though the articles don’t exist.)

On the clinical level, there are also risks involved with screenings and diagnostic tests, which could be exacerbated by AI models that are still fine-tuning their specificity and sensitivity. Tools that incorrectly flag concerns in healthy patients or that provide false comfort to patients who have an issue that testing missed both have the potential to do more harm than good. “That’s why we do a lot of validation trials on different populations across ethnic groups to make sure these algorithms work robustly, no matter who they’re addressing,” Kapa says.

Another, lesser concern is the sheer amount of information overload that will be facing health care providers with the rise of biometric data that people can collect on themselves. AI-driven diagnostic tools are only as useful as the quality and compatibility of the data they have to work with. It’s crucial that the incoming data is accurate, standardized, and interoperable.

If a patient visits a clinic with an Apple Watch, a continuous glucose monitor, and an Oura Ring, and they want help interpreting all that data (which isn’t synced up to their electronic health record), a clinician may be at a loss for how to help. “Patients are coming in with mountains of data, and the interoperability piece of it becomes really challenging,” Jelinek says. “And a lot of these products aren’t FDA-approved, so it’s hard to assess the quality of the data.”

Looking Ahead

Still, it’s hard to miss the general sense of optimism and possibility among health care leaders thinking about how AI stands to revolutionize the patient experience.

Software algorithms loaded onto a smartphone can make sophisticated cardiac testing available to a patient in rural Mississippi, who previously would have had to travel eight hours to the nearest major hospital. Emotionally enabled behavioral health bots will consistently provide timely, empathetic, personalized support and help address root causes before they lead to disease. Physicians will be relieved of administrative tasks that drain the attention they have available for patients.

All this will be in service of supporting the ability of people to take care of people, Jelinek says. “There will always need to be human interaction to care for people. AI has a profound opportunity to help with that. We’re looking to leverage technologies to make providers better at what they do and allow them to focus on the things that matter.”

Click to view the digital edition of our Top Docs issue!