Canada’s largest network of research hospitals has appointed a chief artificial intelligence scientist to harness promising technology that has the potential to speed up diagnoses, improve and personalize patient care and shorten recovery times.
Bo Wang, whose expertise at the Toronto-based University Health Network includes machine learning and computational biology, is stepping into the role after the launch of UHN’s AI Hub earlier this year. The hospital network says the hub brings together doctors and researchers who work with AI in areas including cancer and cardiovascular disease.
Wang will lead the research into how AI can use vast amounts of anonymized patient data collected from the Toronto area’s diverse population to improve care. He said some AI applications he hopes to explore include development of personalized treatment plans and automated generation of clinical notes.
“The goal is to promote adoption of AI in health care,” Wang said in an interview. “We have lots of research but adoption is quite rare, and I want to change that.”
UHN is not alone in its exploration of AI uses in health care. Other hospitals across Canada have been using the technology in limited ways, such as for analyzing the results of medical scans, with oversight from radiologists.
The ultimate goal is to create individualized treatment plans by having AI analyze vast amounts of information and identify patterns based on everything from genetic data to patients’ symptoms, lab results and medications.
Wang said UHN would work with private companies to integrate their AI solutions into clinical practice, following approval of those technologies by Health Canada.
As a founding member of a Mayo Clinic data network in the United States, UHN would also have access to data sets from other countries including Israel and Brazil, he said.
Wang, who is also a faculty member at Toronto’s Vector Institute, which specializes in AI, was one of the main developers of a demo model called Clinical Camel, which was trained on data from thousands of anonymized UHN medical records. It can summarize long conversations between doctors and patients into clinical notes within seconds, he said.
Health-care providers must approve the notes and they can also add information about a patient’s mood or emotional state, Wang added. Doctors can also ask the chatbot questions about symptoms of certain diseases and diagnoses to inform their patient care decisions.
However, the so-called generative AI model, still under development with researchers from the University of Toronto and McGill University in Montreal, needs improvements to make it more reliable. And Health Canada would have to approve the software to ensure accuracy and safety so it does not make wrong predictions about a diagnosis, said Wang, adding the regulator would also have to be satisfied that patient privacy is protected.
Various companies are also developing similar language models to record consultations between doctors and patients and increase efficiency.
“It’s not happening anywhere yet but we see lots of demos, lots of announcements from big corporations like Microsoft,” Wang said of the technology.
UHN says it hopes to expand on the narrow AI applications already being used at its hospitals. At Princess Margaret Cancer Centre, for example, radiation treatment times have been slashed by nearly half in some cases, based on a predictive model built from UHN data on patient recovery after treatment, response to certain drugs and survival times, Wang said.
“This AI model can automatically decide what the optimized dose is for each radiation (treatment) for this particular patient and what’s the time span between different radiation therapies. So, that maximizes your chance of survival and maximizes your post-treatment recovery,” he said.
“The wait time for the patient is smaller, the radiation exposure to the patient is smaller, without sacrificing the treatment’s effectiveness. We are looking at improvements of almost 40 to 50 per cent in radiation.”
Brad Wouters, UHN’s executive vice-president of science and research, said that while AI presents an enormous opportunity in health care, there are “obvious concerns” related to patient privacy and safeguarding data.
That’s why UHN will not share even its anonymized data with the Mayo network or vice versa, he said.
“What’s shared, actually, are the algorithms and tools that train on the data,” he said. “The data never actually leaves or is mixed or under the auspices of any other organization.”