The COVID-19 pandemic has dramatically altered the shape of healthcare around the country, including when it comes to priorities around incorporating artificial intelligence and machine learning into a system’s landscape.
But the need for clinical research has not slackened amid the crisis. In fact, the pandemic demonstrates how solid foundational work to implement AI and ML into workflows can be beneficial – and even crucial – both for solving short-term, urgent issues and for planning longer-term strategies.
Experts at the Mayo Clinic say AI and ML are powerful tools for clinical research and care.
“Rather than thinking of AI/ML in medicine as ‘man vs. machine,’ we like to think of it as (wo)man with machine,” said Dr. Tufia Haddad, a medical oncologist at the Mayo Clinic, in an interview with Healthcare IT News.
“The possibilities are endless in what we can accomplish together with AI/ML in medicine and patient care. Think of the data shackled in our EHRs, labs, radiology and pathology images. The breadth of data contained within a single patient genome and microbiome. And all the data our patients generate each day that we do not yet routinely collect,” she continued.
“I am so impatient for AI to become more of a reality in my day-to-day practice,” added Mayo Clinic Radiation Oncologist Dr. Nadia Laack.
“Every day when I review radiation plans for babies with tumors in their little bodies, I feel like I am walking through a field of land mines. Did I push hard enough to get radiation dose out of their brain, but not so hard that I am undertreating their tumor? Did I keep their kidneys safe so they don’t end up on dialysis as an adult, but not risk cheating on covering the tumor which will kill them this year?”
Laack explained that each organ or body structure has a unique sensitivity to radiation dependent on multiple factors, and that each radiation plan depends on a nearly 100-point checklist. “For example, there might be three different measures for what a safe dose to the spinal cord is, and, despite our fancy tools, we make compromises on almost every plan,” she said.
“I am anxiously awaiting the day when I can rely on AI/ML to help make sure we get to the best plan possible for each patient and double check all my work so I can sleep better at night,” she continued.
In a complimentary HIMSS Learning Center presentation set to air October 8, Haddad, Laack and Google Health UK lead and medical director Dr. Dominic King will share findings around the potential for artificial intelligence in accelerating clinical research and in furthering patient care.
“The majority of AI/ML work that is ongoing at Mayo, and in medicine more broadly today, remains in the realm of research,” said Haddad.
As Haddad explained, many of the applications implemented into Mayo’s practice involve natural language processing, with systems that can recognize and interpret the language in clinical notes, pathology reports and other structured data elements. Some of the systems can assess patients at risk for a medical event or in need of a diagnostic screening; others can leverage NLP and ML to match patients to clinical trials.
“While these help us to better understand our patients, work is still needed to determine if … earlier identification of risk can translate to better clinical outcomes,” Haddad said.
Laack described part of the specific work being done by Mayo researchers using AI and ML, including using it to evaluate FDOPA PET scans of the brain.
“Similar to others who are working in the new field of radiomics, ML algorithms defined features of scans that are not visible to the human eye that can accurately determine the genetic signature of brain tumor,” she continued.
“In patients in whom the pathology test is indeterminate because the sample wasn’t good, or the stain didn’t work very well, we could predict the prognosis of the patient. Similar work has been done by our colleagues in radiology with MRIs and other molecular signatures.”
Laack also noted that for her research in brain tumors, her team has partnered with radiology to use AI and ML to monitor treatment response and automate the contouring of prostate cancer radiation treatment plans.
“Data in radiation oncology is in DICOM format and standard throughout the industry. But sending notations/contours/segments back and forth with our radiology colleagues has been a challenge that we have had to overcome, since most of their efforts did not use DICOM format,” Laack said.
As with previous HIMSS Learning Center presentations, Haddad noted the importance of minimizing disruptions when integrating artificial intelligence into clinical workflows.
“If that’s not feasible, then workflows may need to be redesigned,” Haddad said.
In this way, Haddad pointed out, AI can work in tandem with clinicians – without, of course, replacing them.
“AI is complementary to our strengths – AI/ML can identify patterns in big data sets and quickly/precisely locate knowledge. AI/ML can eliminate bias. And computers can run 24/7 with endless capacity; whereas doctors need to sleep and eat to refuel,” Haddad said. “But only doctors [rather than AI models] can uniquely provide care with compassion, common sense and morals.”
Still, she said, “there is incredible resistance to bringing technology into the medical practice when so many clinicians blame technology on ruining the profession and patient-provider relationship.”
“Sometimes the AI/ML development is 100 times easier than the change management required to successfully implement these models!” she added.
Although AI and ML can be useful for combating human bias, Haddad said it’s critical that the systems be trained in ways that will represent the patient community equitably.
“Data sets selected for training and validation, as well as final testing, should aim to be inclusive of the population to which the model will be applied,” she said.
“If there are limitations in these data sets, then collaboration should be considered to broaden them during model development – or final models may need to be revalidated in new patient cohorts,” she continued.
Mayo Clinic researchers also advocate for transparency about this information as models are disseminated.
As far as the future, the researchers pointed to the potential for AI and ML to scale the processes of automated segmentation of normal organs and teleradiotherapy.
“But that is only the beginning,” Laack said. “What we really need is AI/ML to help us review the patient’s pathology, radiology images and genomic information and help us decide who even needs radiation; should their radiation target nearby lymph nodes or not; or who is doomed to relapse with standard therapy and should be considering something novel like protons or carbon particle therapy; or who needs a targeted chemotherapy agent with their radiation.”
“The answers to our important clinical research and patient care questions may be closer than we think. We can radically transform the conduct of medical research and delivery of patient care with AI/ML,” Haddad said.
“Views from the Top: An AI boost for clinical research and care” is scheduled for Thursday, October 8, at 1 p.m. CT on the HIMSS Learning Center. Click here to register.
Kat Jercich is senior editor of Healthcare IT News.
Email: [email protected]
Healthcare IT News is a HIMSS Media publication.
Source: Read Full Article