Princess Margaret’s researchers have tested the impact of the deployment of Artificial Intelligence (AI) for radiation cancer therapy in a real-world clinical setting in a single study with doctors and their patients.
A team of researchers directly compared the doctors ’assessments radiation treatments generated by an automatic AI (ML) learning algorithm to conventional human-generated radiation treatments.
They found that in most of the 100 patients studied, treatments generated with ML were considered clinically acceptable for physician treatments by patients.
Overall, 89% of ML-generated treatments were considered clinically acceptable for treatments and 72% were selected over human-generated treatments in frontal comparisons with conventional human-generated treatments.
In addition, the ML radiation treatment process was 60% faster than the conventional human-led process, reducing the overall time from 118 hours to 47 hours. In the long run, this could lead to substantial cost savings through improved efficiency, while improving the quality of clinical care, a rare benefit.
The study also has broader implications for AI in medicine.
Although ML treatments were preferably overwhelming when evaluated outside the clinical setting, as is the case in most scientific work, the physician’s preferences for ML-generated treatments changed when the chosen treatment, ML or human-generated, would be used to treat the patient.
In this situation, the number of ML treatments selected for patient treatment was significantly reduced, issuing a cautionary note for teams considering considering not properly validated AI systems.
Results of the study team led by the doctors. They are published by Chris McIntosh, Leigh Conroy, Ale Berlin and Tom Purdie Nature medicine, June 3, 2021.
“We’ve shown that AI can be better than the human criterion for treating radiation therapy with curative intent. In fact, it’s amazing that it works so well,” says Dr. McIntosh, scientist at the Peter Munk Cardiac Center, Techna Institute, and president of Medical Imaging and AI in the Joint Department of Medical Imaging and University of Toronto.
“An important finding is what happens when you actually deploy it in a clinical setting compared to a simulated one.”
Adds Dr. Purdie, a physician, Princess Margaret Cancer Center: “There has been a lot of excitement generated by AI in the lab and the assumption is that these results will translate directly into a clinical setting. so they don’t.
“Once you put ML-generated treatments in the hands of people who trust him to make real clinical decisions about his patients, that ML preference may fall. There may be a disconnect between what happens in a lab-type environment. and a clinic one. ” Dr. Purdie is also an associate professor in the Department of Radiation Oncology at the University of Toronto.
In the study, radiation oncologists were asked to evaluate two different radiation treatments, either ML or human-generated, with the same standardized criteria in two groups of patients who were similar in demographics and disease characteristics.
The difference was that a group of patients had already received treatment, so the comparison was a “simulated” exercise. The second group of patients was about to begin treatment with radiotherapy, so that if the treatments generated by AI were considered superior and preferable to their counterparts, they would be used in the actual treatments.
Oncologists were unaware of what radiation treatment was designed by a human or a machine. Specialist radiotherapists created human-generated treatments individually for each patient according to the normal protocol. In contrast, each ML treatment was developed using a computer algorithm consisting of a database of high-quality radiotherapy plans and peer-reviewed peers of 99 patients previously treated for prostate cancer in Princess Margaret.
For each new patient, the ML algorithm automatically identifies the most similar patients in the database, using similarity metrics learned from thousands of patient imaging functions and delimited target and healthy organs that are standard part of the process. of radiotherapy treatment. Complete treatment for a new patient is deduced from the most similar patients in the database, according to the ML model.
Although the treatments generated by ML were highly valued in both groups of patients, the results of the pretreatment group diverged from the post-treatment group.
In the group of patients who had already received treatment, the number of ML-generated treatments selected on humans was 83%. This fell to 61% for those selected specifically for treatment, prior to treatment.
“In this study, we say researchers need to pay attention to a clinical setting,” says Dr. Purdie. “If physicians believe that patient care is at stake, this may influence their judgment, even though ML treatments are thoroughly evaluated and validated.”
Dr. Conroy, Princess Margaret’s medical physicist, notes that following the success of the study, ML-generated treatments are now used to treat most patients with Princess Margaret’s prostate cancer.
This success is due to careful planning, judicious gradual integration into the clinical environment, and the participation of many stakeholders throughout the process of establishing a robust ML program, he explains, adding that the program is refined. constantly, that oncologists are continually consulted and give feedback, and the results of how ML treatments reflect clinical accuracy are shared with them.
“We were very systematic in the way we integrated it into Princess Margaret’s clinic,” says Dr. Berlin, Princess Margaret’s clinical scientist and radiation oncologist. “It took about six months to build this new software, but it took more than two years to get everyone involved and feel comfortable with the process. Vision, audacity and tenacity are key ingredients, and the Princess Margaret we are fortunate to have leaders from all disciplines who embody these attributes. ” Dr. Berlin is also an adjunct professor in the Department of Radiation Oncology at the University of Toronto.
The success of launching a study of this caliber was largely based on the commitment of the entire Princess Margaret genitourinary radiation cancer group, including radiation oncologists, medical physicists and radiotherapists. It was a great multidisciplinary team effort with everyone’s ultimate goal of improving radiation cancer treatment for Princess Margaret’s patients.
The team is also expanding its work to other places where cancer is developing, including lung and beast cancer, with the goal of reducing cardiotoxicity, a possible side effect of treatment.
Clinical integration of machine learning for the treatment of radiation with curative intent in patients with prostate cancer, Nature medicine (2021). DOI: 10.1038 / s41591-021-01359-w , www.nature.com/articles/s41591-021-01359-w
University Health Network
Citation: AI surpasses humans in creating cancer treatments, but do doctors trust them? (2021, June 3), retrieved June 3, 2021 at https://medicalxpress.com/news/2021-06-ai-outperforms-humans-cancer-treatments.html
This document is subject to copyright. Apart from any fair treatment for private study or research purposes, no part may be reproduced without written permission. Content is provided for informational purposes only.