Virgo Surgical Video Solutions’s collaboration with the Rajpurkar Lab marks a significant shift in how AI will be built for gastroenterology—not as isolated algorithms, but as foundation models trained on massive, multimodal datasets.
At the core of this effort is scale: Virgo is contributing over one million endoscopy videos (within a dataset of millions of procedures), enabling the development of models that go beyond narrow tasks like polyp detection. The goal is to create systems capable of disease detection, risk scoring, treatment prediction, and outcome forecasting—all from routine endoscopic procedures.
What’s particularly important is the move from frame-based AI to video-level understanding. Traditional models analyze static images; foundation models trained on full procedural videos can capture temporal context—how lesions evolve during a procedure, how anatomy is navigated, and how subtle patterns unfold over time. This unlocks a new layer of clinical intelligence that current AI tools largely miss.