Virgo

Market position

Virgo describes itself as the leading cloud video and AI platform for endoscopic medicine and benchmarks EndoDINO against endoscopy foundation models from Roche (Etro), Johnson & Johnson (ArgesFM), EndoFM, GastroNet-5M and DovaVision, reporting larger pretraining scale and higher benchmark scores than these comparators. The company was included in Elion Health's Foundation Model Platforms and APIs market map (January 2026) and in a Navistrat Analytics AI in Endoscopy market report (2026).

Virgo's stated differentiation is the scale and diversity of its endoscopy pretraining dataset, state-of-the-art benchmark results using frozen features rather than task-specific supervised training, a single reusable representation layer applicable across scoring, detection, prediction and biomarker discovery tasks, and an installed base of video capture infrastructure across US endoscopy sites that feeds both clinical and trial use cases.

Technology

The company's technology centers on EndoDINO, a ViT-g/14 vision transformer trained with self-supervised learning on unlabeled endoscopy imagery. The pipeline captures raw endoscopy procedure video, extracts, deduplicates and curates frames, pretrains at population scale, and outputs a reusable embedding for downstream tasks. Reported pretraining scale is 130,037 videos and 3.5 billion frames, described as 57x more frames than the next-largest cited dataset. On HyperKvasir three-class Mayo endoscopic scoring with a frozen backbone and linear probe, EndoDINO ViT-g/14 reports macro F1 of 0.748, above cited comparator figures for DINOv2 ViT-g/14 (0.735), supervised DenseNet (0.729), Roche's Etro (0.706) and EndoFM (0.699). The model also produces four-class Mayo endoscopic scores from frame-level predictions aggregated per procedure. Underlying data is drawn from 148 US medical centers and more than 3 million US procedures, with 46.6% non-White representation and a reported Shannon diversity index of 0.713.

Go-to-market

Virgo distributes its video capture and management platform directly to provider organizations, emphasizing scalable multi-facility deployment and EHR integration, and partners with pharmaceutical trial sponsors for recruitment acceleration. The company builds visibility through peer-reviewed and preprint publication (EndoDINO on arXiv), conference presentations at DDW and UEGW, press releases, industry podcasts, and a research contact channel on its website.

Academic medical centers, integrated health systems and private practice provider networks; gastroenterologists and pulmonologists; pharmaceutical clinical trial sponsors and their participating trial sites.

Geography

Operations and data are centered in the United States, with underlying data drawn from 148 US medical centers; press materials reference Carlsbad, California, and the company portal is operated by Virgo Surgical Video Solutions, Inc. The company has stated plans to expand distribution both domestically and internationally.

History

A Gust company profile lists Virgo as founded in October 2016 in San Francisco as a hardware and software platform for automatic recording, storage and use of surgical video, with the stated aims of improving operating room quality control, standardizing HIPAA-compliant surgical video recording, and improving access to surgical video for training and research. The company subsequently focused on endoscopy, launching VirgoTrials for clinical trial recruitment in 2021 and closing a Series A led by FCA Venture Partners. From January 2025 it began publicizing its EndoDINO foundation model, including an arXiv paper (January 2025) and the EndoML platform launches announced in January and May 2025.

Risks & controversies

Virgo has published a notice that individuals have impersonated Virgo recruiters, stating that the company emails only from @virgosvs.com, conducts interviews on Google Meet, and never asks candidates to pay for equipment or supply banking or tax details. Several headline performance claims, including the UNIFI cost and timeline savings, are retrospective modelling presented by the company rather than prospective validation.