In this webinar, host Ravi Parikh, MD, MPP, FACP, of Winship Cancer Institute, Emory University speaks with Danielle S. Bitterman, MD, of Harvard Medical School, and Drew Williamson, MD, of Emory University School of Medicine, about how AI is being used across the entire oncology care continuum to support more coordinated patient care.
Dr. Williamson explains that the pathology field has been focused on the necessary digitization of pathology slides to enable more widespread adoption of AI. In terms of pathology research, AI algorithms are being developed and explored for single-use biomarker-led use cases up through larger foundation models.
Radiation oncologists, Dr. Bitterman explains, are already actively using different models in the clinic, most commonly for auto-segmentation or auto-contouring as part of radiation treatment planning. Research has been focused on response prediction and treatment decision support.
In radiology, AI has already demonstrated benefit in supporting mammography for breast cancer detection, as was seen in the randomized, controlled MASAI trial. In research settings, there’s increasing interest in AI-generated radiology reports.
In clinical practice, AI algorithms have been used for quantifying immunohistochemical staining and some institutions have integrated AI into their workflow for review of prostate cancer biopsies and determining Gleason scores via Paige Prostate, Ibex Prostate, and other companies. Dr. Williamson explains that at Emory, his colleague Andrew Janowczyk, PhD, MS, has developed an AI tool for detecting H pylori in gastrointestinal biopsies.
Research in computational pathology has also explored the detection of molecular alterations from histology images, such as the detection of EGFR mutations in lung cancer slides. Dr. Williamson also expects more AI pathology predictive models to enter the market to expand the possibility of what pathologists can find in images. Dr. Parikh notes that AI-supported biomarkers have already been accepted into NCCN guidelines, such as the ArteraAI Prostate Test, but these still have not been widely adopted by practices yet, despite the supporting evidence.
Natural language processing and large language models have been discussed most widely and implemented in terms of ambient documentation, which has already been adopted in several health-care practices, with some trials demonstrating efficiency benefits. These AI approaches are also considered for automated data extraction from reports, clinical decision support tools, and clinical trial matching.
In the clinical trial operation space, Dr. Bitterman explains that AI can be useful from beginning to end, from setting up the trial all the way through to acting as a navigator for patients and answering their questions.
There are also opportunities for AI to be used more for molecular tumor boards to keep the group up to date on guidelines, to take less preparation time, and to ensure no important information is skipped.
Going forward, Dr. Bitterman says that monitoring of AI models and tools is a significant gap in research, and Dr. Williamson says that greater data sharing and multidisciplinary collaboration is needed for even more positive impacts of AI models in the oncology space.
Listen to the full webinar for more explanations and examples of AI uses across the oncology care continuum.