Mayo Clinic and Google Health have announced they will use artificial intelligence to improve radiation therapy planning for cancer care.
The project is the first initiative in a 10-year strategic partnership between the Rochester, Minnesota-based hospital and the tech giant, announced in September 2019.
Radiation oncologists, medical physicists, dosimetrists, and service design professionals from Mayo Clinic will work together with Google Health’s AI, medical image segmentation, and user experience design experts. In addition to AI, the two companies plan to use Google Cloud and data analytics to advance the diagnosis and treatment of diseases.
“We’ve been working with Mayo Clinic now for over a year to try and bring some of our expertise around medical imaging analysis with artificial intelligence or machine learning to try to help tackle this task of segmenting or contouring scans in radiotherapy,” Cían Hughes, informatics lead at Google Health, told Fierce Healthcare.
Want to publish your own articles on DistilINFO Publications?
Send us an email, we will get in touch with you.
During the project’s first step, announced Wednesday, Mayo Clinic and Google Health will create and validate an algorithm that can automate healthy tissue and organs from tumors. Developed using de-identified data, the algorithm will boost patient outcomes and reduce the time it takes to plan radiation treatment, the companies said.
Regulatory experts at Mayo and Google also are exploring potential approaches to get approval from the U.S. Food and Drug Administration (FDA) for a medical device that incorporates these algorithms, Hughes said during a press conference Wednesday.
“Two key questions will be determining what class the medical device would be and the accuracy of those algorithms and the level of human intervention required. We think the results of this initial research will significantly influence all of those decisions around the FDA and medical devices,” Hughes said.
About 50% of cancer patients seek treatment with radiotherapy during their course of diagnosis, according to Nadia Laack, M.D., chair of the department of radiation oncology at Mayo Clinic and a principal investigator on the project.
In the past, radiation planning would take about a half-hour, and medical professionals would only need to calculate a radiation dose based on measuring a patient’s width, said Laack in an interview. Now the treatment is more labor-intensive and takes additional time because techniques are more sophisticated with additional detail.
“It takes computer algorithms, computer models to estimate and calculate a dose now with the modern radiotherapy technique,” Laack said. “It takes a lot longer than it used to because there are so many more complexities in the way that the dose comes into the patient and is delivered.”
Machine learning algorithms will lead to improved radiotherapy planning and better patient outcomes, according to Laack.
“You’ll be able to give it more information to accurately mark out where the tumor is and the lower the toxicity that the patient will have,” Laack said. She added that less toxicity means “less dry mouth, reduced risk of spinal cord injury, and fewer heart problems.”
Google Health and Mayo Clinic are using a deep learning algorithm called U-Net that provides the detail of tiny pixels that are essential when studying scans of tumors, according to Hughes.
“It’s particularly good because it preserves the resolution in the output,” Hughes said.
In addition, the automated contouring performed using ML could help avoid organ damage from radiation, Hughes noted. It will also allow physicians to spend more time with patients and their families.
User experience design experts at Google are working with a team of Mayo’s service design professionals to look at how these algorithms can be incorporated into the clinical workflow, Hughes said during the press conference. “We are deeply researching and understanding the human-computer interaction to ensure that we are decreasing rather than increasing the possibilities for errors and variabilities in care.”
At first, medical researchers will focus on radiation treatment for cancers in the head and neck areas. Going forward, Mayo Clinic and Google Health plan to train similar algorithms that can help them plan treatment for other areas of the body with cancer risk, such as the breast and prostate, Hughes said.
“We’re hoping that the results are promising and that we’ll be able to disseminate and translate this knowledge throughout the rest of the body,” Laack added.
The AI project is in the research stages, and the algorithms are not being used for real-life treatment of patients today, according to Hughes.
“However, we’re hopeful that if this research is successful and if we get good results from that modeling work, we will be able to translate this into tools for use in the care of patients,” Hughes said. He added that due to the care that Google Health and Mayo Clinic take as far as testing the safety of the tools, “that’s probably a journey that will take us years.”
Source: Fierce Healthcare