AI is better than humans in developing cancer treatments, do doctors trust it?

The impact of the deployment of artificial intelligence (AI) on radiation cancer treatment in the real-world clinical environment has been tested by Princess Margaret’s researchers in a unique study of doctors and their patients.

The research team directly compared doctors’ evaluations. radiation Therapies generated by AI machine learning (ML) algorithms replace traditional human-generated radiation therapy.

They are the majority of 100 people patient In this study, treatments generated using ML were considered clinically acceptable as treatments for patients by physicians.

Overall, 89% of the treatments produced by ML were considered clinically acceptable for treatment, and 72% were selected over human-generated treatments compared to traditional human-generated treatments. ..

In addition, the ML radiation therapy process was 60% faster than traditional human-led processes, reducing overall time from 118 hours to 47 hours. In the long run, this may mean significant cost savings through improved efficiency, while at the same time offering the rare benefit of improving the quality of clinical care.

This study also has a wide range of implications for AI in healthcare.

ML treatments were overwhelmingly preferred when evaluated outside the clinical environment, as done in most scientific studies, but physicians’ preference for treatments generated by ML is ML or humans. The selected treatment generated by was modified when used to treat the patient.

In that situation, the number of ML treatments selected to treat patients was significantly reduced, warning teams considering deploying an under-validated AI system.

Results of a research team led by Dr. Chris McIntosh, Lee Conroy, Yale Berlin, Tom Purdy Natural medicine, June 3, 2021.

“We have shown that AI is better than human judgment in curative radiation therapy. In fact, it’s amazing how well AI works,” he said. Dr. McIntosh, a scientist at the Petermank Heart Center at the Techna Institute, said: Co-department of the Medical Imaging Department at the University of Toronto and Chief of Medical Imaging and AI.

“The main finding is what happens when it is actually deployed in a clinical environment compared to what is simulated.”

Dr. Purdy, a medical physicist at the Princess Margaret Cancer Center, adds:

“In a lab-based setting, we can reduce our preference for ML by providing ML-generated therapies in the hands of those who rely on ML to make real clinical decisions about their patients. What’s happening, what’s happening clinically? “Dr. Purdy is also an associate professor of radiation oncology at the University of Toronto.

In this study, the treating radiation oncologist evaluated two different radiation therapies (ML or human-generated) on the same standardized criteria and identified two patient groups with similar demographics and disease characteristics. It is targeted. I asked.

The difference is that the comparison is a “simulated” exercise because one group of patients has already been treated. A second group of patients is about to start radiation therapy and will be used in the actual treatment if the AI-generated treatment is determined to be superior or preferable to human treatment.

Oncologists were unaware of which radiation therapy was designed by humans or by machines. Artificial treatments were created for each patient according to the usual protocols of professional radiation therapists. In contrast, each ML treatment was developed by a computer algorithm trained in a high-quality, peer-reviewed database of radiation therapy schemes from 99 patients treated for prostate cancer at Princess Margaret.

For each new patient, the ML algorithm uses thousands of features taken from patient images and similarity indicators learned from contoured targets and healthy organs that are a standard part of the radiation therapy process. , Automatically identify the most similar patients in the database. Complete treatment of new patients is inferred from the most similar patients in the database according to the ML model.

The treatment generated by ML was highly valued in both patient groups, but the results in the pretreatment group differed from those in the posttreatment group.

In a group of already treated patients, 83% of treatments for ML generation were selected over human treatments. This was reduced to 61% prior to treatment for patients specially selected for treatment.

“In this study, researchers say they need to pay attention to the clinical environment,” says Dr. Purdy. “If a physician feels that his or her care is at risk, even if the ML treatment is thoroughly evaluated and validated, it can affect the physician’s judgment.”

Following a very successful study, Princess Margaret’s medical physicist, Dr. Conroy, said the treatments produced by ML are currently used to treat the majority of Princess Margaret’s prostate cancer patients. It was. I’m pointing out.

Her success is due to careful planning, wise step-by-step integration into the clinical environment, and the involvement of many stakeholders throughout the process of establishing a robust ML program, and the program is always sophisticated. .. Oncologists are continually consulted and provide feedback. And the results of how well ML treatment reflects clinical accuracy are shared with them.

“The way we integrated this into Princess Margaret’s clinic was very systematic,” said Dr. Berlin, a clinical scientist and radiation oncologist at Princess Margaret. “It took us about six months to build this new software, but it took more than two years for everyone to get involved in the process. Vision, boldness, and tenacity are key factors. Fortunately for Princess Margaret. Best of all, she is the overall leader in the field that embodies these attributes. “Dr. Berlin is also an assistant professor at the Department of Radiation Oncology, University of Toronto.

The successful initiation of this excellent study relied heavily on the efforts of Princess Margaret’s entire urogenital radiation cancer group, including radiation oncologists, medical physicists, and radiation therapists. This was the work of a large interdisciplinary team with the ultimate goal of improving radiation cancer treatment for Princess Margaret’s patients.

The team is expanding its research to other cancer sites, such as lung cancer and veterinary cancer, with the goal of reducing the side effects of treatment, cardiotoxicity.