An international, multidisciplinary team of neuro-oncology researchers and physicians has developed new clinical practice recommendations for the use of artificial intelligence (AI) methods to more accurately diagnose, monitor, and treat patients with brain cancer. The guidelines were described in a policy review published by Bakas et al in The Lancet Oncology.
Background
The standard of care in neuro-oncology currently involves individual radiologists measuring tumor size to inform treatment options. However, this strategy is not ideal because the assessment is often subjective; each radiologist can interpret imaging scans differently, leading to treatment approaches that fluctuate based on who views the scans.
Predictive, prognostic, and diagnostic AI models and solutions are becoming available for health-care practitioners, but how they are used may vary widely at different institutions.
“Thanks to new technology, there are ways to use AI to help assess whether a tumor is progressing or is stable,” detailed senior author Raymond Y. Huang, MD, PhD, Associate Professor at Harvard Medical School and Chief of the Neuroradiology Division at Brigham and Women’s Hospital. “However, there needs to be a standardized way to use AI to accurately diagnose and treat patients,” he added.
“We can use AI to look at images of the tumors more objectively,” highlighted lead author Spyridon Bakas, PhD, the Joshua Edwards Associate Professor in Pathology and Laboratory Medicine and Director of the Division of Computational Pathology at the Indiana University School of Medicine as well as a researcher at the Indiana University Melvin and Bren Simon Comprehensive Cancer Center. “AI programs can help determine quickly what type of disease it is, what subtype of tumor, and what particular grade it is in addition to helping track the progress of a lesion during treatment,” he explained.
Guideline Overview
The current review is one of two companion policy reviews conducted on behalf of the Response Assessment in Neuro-Oncology cooperative group, a collaboration of international experts developing standardized criteria for evaluating treatment response in clinical trials for brain cancer.
In this policy review, the researchers established guidelines for the standardization, validation, and good clinical practice of AI in neuro-oncology. The new guidelines cover the use of software that had been developed with large and diverse cohorts of patient data; assurance that AI models for defining a tumor follow World Health Organization criteria; and considerations for how the tumor images are obtained, processed, and segmented prior to analysis.
The researchers emphasized that the new guidelines may provide a much-needed update to the current standard of care.
“These guidelines are critical for ensuring that AI tools developed in the [United Kingdom] and beyond meet rigorous standards and improve patient outcomes,” underscored co-author Thomas Booth, MD, PhD, of King’s College London. “With these recommendations, we can move toward more accurate, standardized AI applications that benefit both clinicians and patients … internationally,” he continued.
Because AI is still new, the recommendations were among the first in the world regarding its proper use in cancer care, and further study is necessary.
“It is important that we continue our study of these AI models on large, diverse patient populations to continue extending our understanding of disease and improving the way we use them,” Dr. Bakas concluded.
Disclosure: The research was supported by the National Institutes of Health, the U.S. Department of Veterans Affairs, the U.S. Department of Defense, the Dana Foundation David Mahoney Neuroimaging Program, the V Foundation for Cancer Research, the European Union’s European Research Council, and the Deutsche Forschungsgemeinschaft. For full disclosures of the guideline authors, visit thelancet.com.
