Artificial intelligence (AI) prescriptions for radiation therapy are “clinically acceptable” in the majority of cases. To say it is a study conducted at the Princess Margaret Cancer Center in Toronto, Canada, by a team of radiotherapists oncologists, medical physicists and artificial intelligence experts: doctors and researchers have evaluated the ability of machine learning systems to develop radiotherapy protocols for 100 patients with prostate cancer in daily clinical practice. Simplifying a bit, a radiotherapy treatment plan is an irradiation program that defines the dose of radiation, the volumes to be hit and the boundaries to be respected between healthy and diseased tissues, and is normally developed by medical physicists on the basis of images (tac, resonances, etc.) and the indications of the oncologist. The conclusion of the research is that the protocols developed by the AI work. But also that it is important to maintain the control systems that only human intelligence can and must put in place, and that it is necessary to promote an attitude of trust among doctors towards these new tools.
ASCO 2021
Prostate cancer, new hope from targeted radiotherapy
by Tiziana Moriconi
Research
The studio captured the cover of Nature Medicine. For each of the 100 patients, on the basis of the images provided by the CT scans, two radiotherapy plans were developed: one by the algorithm and one by medical physicists specialized in radiation treatments. Specifically, 50 of these patients had already undergone radiotherapy and the evaluation was done retrospectively (in practice in a situation that we could say simulated), while the other 50 patients had not yet been treated and would have been after the evaluation of the doctors (prospective phase).
ASCO 2021
Prostate, well-tolerated drug that increases survival in cancer at high risk of metastasis
AI “acceptable”
In summary, things went like this: overall, oncologists considered 89% of the plans generated with AI clinically acceptable. But if you look only at the prospective phase, their evaluations of the treatments generated by the algorithm have changed: in this situation, doctors have chosen the treatments obtained with the algorithm much less often than in the retrospective one. In practice: while the treatments provided by AI were very clearly preferred when they were evaluated outside the clinical setting, as is the case in most scientific works, when it came to applying the treatment to the patient, this preference has shrunk. In reality, when choosing between the automated radiotherapy plan and the treatment plan established by the medical physicist, the oncologists did not know which one was and which was the other, but they had to indicate which, according to them, had been generated by the artificial intelligence.
The merit of the study
In medicine, AI technologies are increasingly integrated with traditional methods in the diagnosis and treatment of many pathologies. But more rarely, artificial intelligence reaches the patient’s bed. And this is already a merit of the study. But he is not the only one. This research also provides an opportunity to reflect on increasingly emerging issues in medicine as the techniques that use AI improve. “We have shown that artificial intelligence can be better than human judgment for radiation therapy. In fact, it is surprising that it works so well,” said Chris McIntosh, chair of Medical Imaging and AI at the University of Toronto and first author of the publication. : “An important discovery – added McIntosh – was seeing what happens when it is actually applied in a clinical setting compared to a simulated one.” The study also evaluated how much time the algorithmic procedure would save compared to the human one: 70 hours on average .
Prostate cancer: Exercise reduces depression
by Tina Simoniello
“When you put AI-generated treatment plans in the hands of people who need to make real clinical decisions about their patients, the preference for machine learning technologies can shrink. There can be a mismatch between what happens in a laboratory setting. and in a clinical one, “added Thomas Purdie, medical physicist at the Princess Margaret Cancer Center and associate professor in the Department of Cancer Radiotherapy at the University of Toronto:” Artificial intelligence in the laboratory has generated a lot of excitement and the hypothesis is that results obtained in an experimental setting will translate directly into a clinical setting. But our research warns us that things may be different. Knowing that patient care is at stake could influence doctors’ judgment, even if the treatments decided by the AI are carefully evaluated and validated. “
Prostate cancer, Covid has canceled a third of the interventions
A possible integration
At Princess Margaret, machine learning-generated treatments are now being used in the treatment of most prostate cancer patients, given the success of this study, the authors say. “A success – said Leigh Conroy, medical physicist – which is due to careful planning, a gradual integration into the clinical environment and the involvement of many interested parties in the process of developing a solid machine learning program”. A program that is constantly being refined and in the course of which oncologists are continuously consulted and continuously provide feedback. And the results obtained on clinical accuracy are always shared.
.