GE HealthCare announced on September 23, 2026 that it is expanding a research collaboration with Mass General Brigham to explore whether generative AI can help radiation oncology teams make sense of treatment data that is currently scattered across incompatible hospital systems. The project, announced from Chicago, would feed into GE HealthCare’s existing Intelligent Radiation Therapy (iRT) software, an AI-supported platform the company already sells for managing radiation therapy workflows.
The Problem: A Patient’s Cancer Treatment Record Is Scattered Everywhere
Radiation therapy planning has long suffered from fragmentation. A single patient’s treatment journey — imaging scans, clinical notes, dose plans, physician assessments — typically lives across multiple systems that rarely talk to each other, forcing care teams to manually hunt down information before they can make decisions. According to GE HealthCare, importing treatment data into a planning workflow can take up to 17 manual steps, and a study the company cites of more than 500 radiation oncology cases found more than 44 distinct variations in how the treatment-planning workflow was actually carried out from case to case — a level of inconsistency that complicates efforts to standardize care or spot errors. The stakes are sizable: per the American Society for Radiation Oncology’s 2023 Radiation Oncology Case Rate Report, nearly 60% of people diagnosed with cancer receive radiation therapy at some point in their treatment.
What the Generative AI Is Meant to Do
Rather than interpreting images or making treatment decisions itself, the generative AI tool under development is designed to combine structured clinical data with unstructured information — clinical notes, scanned images, prior treatment plans — and let care teams simply ask questions to surface relevant details about a specific patient’s treatment history. The goal, according to GE HealthCare, is to give clinicians a summarized view that helps them navigate complex, multi-system treatment data without manually digging through each source, and potentially to extend those capabilities into other iRT-supported workflows involving MR, CT, and theranostics imaging down the line. The company is careful to frame this as exploratory research rather than a finished, commercially available product.
Building on an Existing Track Record
This isn’t GE HealthCare and Mass General Brigham’s first collaboration. Software originally developed at Massachusetts General Hospital, a founding member of Mass General Brigham, was previously incorporated into iRT as part of an earlier partnership. That earlier work is already credited with tangible results: GE HealthCare says the technology helped reduce the time from patient intake to the start of radiation treatment from as long as 30 days down to eight days, based on an analysis of more than 11,000 treatment plans at MGH. The company is careful to note that results may vary depending on a hospital’s workflows, patient population, and other site-specific factors — a caveat worth taking seriously given that the 30-to-8-day figure comes from a single academic medical center with considerable informatics resources, not a broad multi-site trial.
Who’s Behind It
John Wolfgang, PhD, a clinical physicist in the Department of Radiation Oncology at Mass General Brigham, is the principal investigator on the research. “We’re excited to advance this work to explore how AI could further support radiation oncology care teams,” he said in the announcement. Sam Kandala, General Manager for Therapy Guidance at GE HealthCare, framed the project as a continuation of solving concrete clinical pain points rather than a speculative AI showcase: “Our iRT solution began by solving a very real clinical problem,” he said, referring to the fragmented-data problem the new generative AI tool is meant to extend that solution into.
The Harder Questions Generative AI in Oncology Still Faces
Using generative AI to summarize clinical information carries risks that are different from, and in some ways harder to manage than, the risks of an image-reading diagnostic algorithm. Large language model-style tools are prone to hallucination — confidently generating plausible-sounding but inaccurate summaries — and in a radiation oncology context, a dropped detail or misattributed dose history isn’t a minor inconvenience, it’s a potential safety issue in a field where treatment errors can cause serious harm. GE HealthCare’s own framing of this as early-stage “exploratory research” rather than a cleared product suggests the company itself is not yet ready to put this specific generative capability in front of clinicians making real treatment decisions, and any eventual commercial version will likely need to clear a higher validation bar than a straightforward image-analysis tool would.
What’s Next
For now, the collaboration remains a research partnership without a defined commercial launch date, and GE HealthCare has not said when, or whether, the generative AI summarization tool will become an actual feature inside iRT. But the direction is clear: having already shown that software can meaningfully cut intake-to-treatment delays at one hospital, GE HealthCare and Mass General Brigham are now betting that generative AI — applied carefully to the administrative chaos around cancer treatment rather than to clinical decision-making itself — can extend those efficiency gains further, and potentially across other imaging-heavy specialties that share the same scattered-data problem.
Photo: Kampus Production / PEXELS via Pexels
