Dr Borgen on the Strengths and Limitations of AI in Oncology Practice

“There is a cautionary note…if you’re using [an AI] platform, [you need to understand] what its pros and cons and…its limitations are.”

Patrick I. Borgen, MD, chair, Department of Surgery, Maimonides Medical Center; head, Maimonides Breast Center, Maimonides Cancer Center, discusses research outlining the current clinical applications and limitations of artificial intelligence (AI) in oncology practice.

At the 41st Annual Miami Breast Cancer Conference in 2024, Anant Madabhushi, PhD, of the Winship Cancer Institute of Emory University, highlighted groundbreaking advancements in AI for cancer image analysis in a keynote address, Borgen begins. Research conducted at Emory University demonstrated that AI image analysis can match or surpass conventional genomic profiling in predicting biological cancer characteristics, he says. This finding positions AI as a disruptive technology with the potential to revolutionize tumor analysis, challenging a multi–billion-dollar industry centered on genomic, proteomic, and expression assays, he explains. The ability of AI to analyze microscopic images at significantly lower costs and faster speeds than humans could make it a transformative tool in oncology, he emphasizes.

However, it is important to consider AI’s current limitations, Borgen noted. At the upcoming 42nd Annual Miami Breast Cancer Conference, Joshua Feinberg, MD, of the Maimonides Medical Center, will present findings from a study evaluating the performance of ChatGPT and Google Gemini on breast cancer board examinations, including the BESAP, SSO, and ASBrS fellowship tests. Both platforms achieved approximately 70% accuracy on these tests, an outcome that raises concerns for clinical decision-making, Borgen states. A 30% error rate highlights the risks of relying on AI-driven web searches for medical guidance, he says. Furthermore, these platforms struggled with tasks involving image analysis, such as interpreting X-rays and pathology slides, he adds. Although AI holds promise for improving cancer diagnostics and treatment planning, this research emphasizes the importance of understanding the strengths and weaknesses of AI tools, Borgen concludes.