How AI Is Changing Cancer Care from Diagnosis to Survivorship

Travis Osterman, from the Vanderbilt-Ingram Cancer Center presented a review of how artificial intelligence (AI) is being used to improve oncology care at the Annual 2026 NCCN Conference. The presentation focused on the current and future applications of AI across the entire cancer care continuum to support more precise and patient-centered care.

Artificial intelligence applications in cancer care

Artificial intelligence tools are being integrated into several stages of cancer care, including prevention, diagnosis, treatment, survivorship and end-of-life care. In the area of prevention, AI is used to generate personalized risk profiles and improve the efficiency of cancer screening programs. For diagnosis, AI technologies assist with quick and accurate radiology image interpretation, efficient tissue acquisition for histopathology and automated analysis of histopathology images. These tools also support accurate diagnosis through the analysis of molecular data.

During treatment, AI is applied to surgical planning, intraoperative navigation and improving surgical efficiency. In radiation treatment, it helps with target volume delineation and treatment planning. For systemic treatment, AI supports personalized treatment selection, accelerated drug discovery and improved monitoring of side effects.

In the survivorship phase, AI assists with enhanced cancer monitoring, surveillance and the creation of personalized care plans. It is also used for psychosocial interventions for cancer survivors. For end-of-life care, AI tools are used for symptom management in palliative care, providing prognostic information in advanced cancer, assisting with decision making and offering enhanced support to families and caregivers.

Results from AI implementation and studies

Specific implementations of AI at Vanderbilt University Medical Center include the use of ambient scribes, surgical planning, infusion scheduling and radiology critical alerts.

Research presented at the conference highlighted several findings regarding AI's impact on clinical skills and learning. A study by Kosmyna et al. published in September 2025 found that the retention of concepts after writing an essay decreased when using large language models (LLMs) compared to using a web search or having no assistance. Another study by Budzyński et al. published in Lancet in October 2025 observed a "deskilling" effect in endoscopy. The study found that continuous exposure to AI-assisted colonoscopy decreased the adenoma detection rate (ADR) when the AI tool was removed.

Additional data showed that large language models have already passed the United States Medical Licensing Examination (USMLE) and specialty board certifications. However, when used as a "predictive decision support intervention" for diagnosis or treatment recommendations, even a high-performing AI may fall short because it carries the same license and malpractice risks as a human while potentially adding to the cognitive burden of the clinician who must assess its recommendations.

Details of the augmented reality surgery study

The presentation included details from a study by Prasad K et al. titled "Augmented-Reality Surgery to Guide Head and Neck Cancer Re-resection: A Feasibility and Accuracy Study," published in Ann Surg Oncol in 2023. This study addressed the difficulty surgeons face when relocating positive margins. The process involved performing ex vivo 3D scanning of surgical specimens. These 3D files were then uploaded to an augmented reality headset. Surgeons could then align the 3D specimen hologram into the surgical defect to guide the re-resection.

Safety and side effects of AI integration

The integration of AI into oncology care involves several safety considerations and potential side effects related to clinical practice. One identified challenge is "AI reliance," where the use of AI tools may lead to a decrease in the retention of medical concepts. As noted in the endoscopy study, a significant side effect of continuous AI assistance is the potential for deskilling, where a clinician's independent ability to detect abnormalities like adenomas decreases when they are no longer using the AI.

Another safety concern is the "medical student paradox" for predictive AI that provides diagnosis or treatment recommendations. While an AI might perform at the level of a "great medical student," it also introduces a cognitive burden for the clinician who must balance "known-unknowns" and "unknown-unknowns" when evaluating the AI's suggestions.

Regulatory updates are being implemented to manage these risks. The FDA's Predetermined Change Control Plan (PCCP), with final guidance issued Aug 18, 2025, aims to control "algorithm drift" without requiring additional FDA approval for every modification. This involves detailed modification protocols and impact assessments. Furthermore, new requirements for electronic health records (EHR) starting Jan 1, 2025, mandate that "source attributes" of AI-driven decision support interventions be exposed in plain language to the user to improve transparency.

References

1. “Augmented-Reality Surgery to Guide Head and Neck Cancer Re-resection: A Feasibility and Accuracy Study” by Prasad K, et al., Ann Surg Oncol.

For more news on cancer updates, research and education, don’t forget to subscribe to CURE®’s newsletters here.