Mayo Clinic and Google Health on Wednesday launched a joint research project to study whether artificial intelligence can automate aspects of radiation therapy planning.
Radiation therapy is common cancer treatment that involves using X-ray, proton or other energy beams to kill cancer cells. A critical part of planning for the therapy is contouring, or segmenting healthy tissue and organs from nearby tumors. Mayo and Google Health officials think contouring could be helped with AI.
Today, that process of identifying and differentiating healthy and cancerous areas is largely manual or "semi-automated," and can take six-plus hours for complex cases, according to Dr. Nadia Laack, chair of Mayo's radiation oncology department in Rochester, Minn., and one of the principal investigators on the research project.
When planning for a case that involves using radiation therapy on a patient's head, for example, a radiation oncologist would draw detailed lines on medical scans around a patient's eyes and spinal cord to ensure radiation beams avoid those areas.
"The computer models that we use (to) help with designing the radiation plans … can't see those organs unless we segment them," Laack said on a call with reporters Wednesday. "All those things must be demarcated so the computers can see them and avoid them."
By applying AI to automate parts of the contouring process, collaborators at Mayo and Google Health hope the research will one day allow care teams to generate radiation dose plans that better protect patients' healthy tissue and organs, treat tumors in a more targeted way and reduce treatment planning times, as well as overall decreasing variability.
The first step of the research project—a five-year collaboration—will involve developing and validating an algorithm that segments healthy structures from tumors, as well as developing associated dosage and treatment plans for patients undergoing radiation therapy. The project initially will focus on treatment for head and neck cancers.
Mayo and Google Health researchers will train the algorithms on medical scans that have already been segmented by physicians, said Dr. Cian Hughes, an informatics lead at Google Health, during the call. The scans are de-identified and from Mayo patients who have consented to having data shared as part of research studies, he added.
Another component of the research project's first phase will involve studying how the AI system would be "safely placed within the clinical workflow," according to Hughes. That part of the project will bring in researchers in service design from Mayo and user experience from Google to assess how human clinicians will interact with the computer program.
The technology won't be used in clinical practice yet since it's being researched, according to Hughes.
Mayo radiation oncologists, medical physicists and other radiation therapy experts will work with teams at Google Health who have researched AI and medical imaging. In 2018, researchers from Google and the University College London Hospitals published a study on scientific preprint website arXiv.org illustrating how their AI system could segment medical scans of patients with head and neck cancers.
Mayo's research project with Google is part of a 10-year partnership that the organizations announced in September of last year, which includes data storage and innovation projects.
Mayo officials have stressed that the health system will control access to patient data under that 10-year partnership; however, they may authorize partners like Google to use data as part of specific projects.
Google has made numerous moves into the healthcare market in recent years, many of which have entangled the company in controversies related to privacy and concerns over data access.
That's included the company's involvement with an online COVID-19 screener launched by one of its sister companies, as well as a massive data-sharing partnership with St. Louis-based hospital giant Ascension. Most recently, Google invested $100 million into telemedicine company Amwell.