Medical devices & diagnostics · Gastrointestinal AI
Docbot
A UC Irvine spinout that turned one gastroenterologist’s decade of colonoscopy images into real time polyp detection AI. It raised $8.5 million, opened a 978 patient randomized trial, never reached FDA clearance, and sold its assets in November 2022.
Ten years of pictures
Dr. William Karnes arrived at UCI Health with one assignment: wipe out colorectal cancer in Orange County. He started with the part he could control, saving images from every colonoscopy performed at his digestive disease center.
Within a year, UC Irvine News reported, the tens of thousands of images he had collected helped the group raise its already good detection rates by 50 percent. Doctors who can see what they are missing start missing less.
Then he ran out of eyes. He found software in Andrew Ninh, an information technology specialist at UCI’s Beall Applied Innovation center who, shortly before his eighteenth birthday, had been hospitalized with a collapsed lung and watched nurses tune out a ward full of alarms that cried wolf. In 2013, by the company’s own account, Ninh, Tyler Dao and Karnes teamed up to build a web front end for the polyp database Karnes had spent ten years assembling. What they layered on top became UltivisionAI.
The number that will not move
Colonoscopy is the only screening test that finds and removes a precancerous lesion in the same sitting. Most interval cancers trace back to polyps that were missed or incompletely removed, and the weak point is the operator: adenoma detection rate varies widely between colonoscopists, and patients whose doctor has a high rate have the lowest subsequent cancer rates.
Computer aided detection, or CADe, was meant to close that gap by drawing a box around suspicious tissue in real time. The evidence has been more stubborn than the pitch. A 2023 review in Clinical Gastroenterology and Hepatology pooled eight nonrandomized studies covering 9,782 patients and found no difference in detection rate between CADe and standard colonoscopy, 44 percent against 38 percent, a risk ratio of 1.11 with a confidence interval of 0.97 to 1.28. In 2025 a multicentre study in Poland, in The Lancet Gastroenterology and Hepatology, reported something more uncomfortable: after three months of routine AI use, the same endoscopists working without AI saw their detection rate fall from 28.4 percent to 22.4 percent.
A database, an accelerator, and a grant
Docbot set up inside UCI Health in 2018 and joined the Wayfinder incubator at UCI Beall Applied Innovation. That June it announced Qualoscopy, pairing polyp detection with automated reporting of quality measures such as prep score, cecal intubation and withdrawal time. In March 2019 it came out of Y Combinator’s Winter 2019 batch, and by May had closed $2.045 million in seed funding from Khosla, Lightbridge Ventures, BOLD Capital Partners, Collaborative Fund, Fenox Venture Capital and three angels. The National Science Foundation added a $225,000 SBIR Phase I award, with Ninh as principal investigator, for Barrett’s esophagus dysplasia detection.
The science was real and it was published. In January 2020 the American Journal of Gastroenterology carried a Docbot paper showing a convolutional network that predicted diminutive polyp pathology with a 97 percent negative predictive value in the rectum and rectosigmoid, clearing the society thresholds for resect and discard. In February 2021 Gastroenterology published work with Eli Lilly in which a recurrent network read 795 full length endoscopy videos from a phase 2 trial, 19.5 million frames in all, and agreed with human central readers on ulcerative colitis severity at a quadratic weighted kappa of 0.844 and 0.855.
The pivotal trial, and then silence
On June 24, 2020, Docbot announced the first subjects enrolled in AIDA, a randomized trial comparing standard high definition colonoscopy with and without UltivisionAI. The release described 978 subjects across six US sites and said the company believed the trial could support a De Novo submission. The registry entry, last updated March 17, 2021, lists nine facilities.
In January 2021 the board brought in an operator. Andrew J. Ritter, who had taken Ritter Pharmaceuticals public and sold it to Qualigen Therapeutics in 2020, became CEO. His appointment was announced in May alongside a $4 million Series A led by Khosla Ventures, bringing the total raised to $8.5 million. Founding CEO Ninh moved to chief strategic officer.
Then the public record stops. AIDA has not been updated since March 2021 and shows no posted results. No 510(k) or De Novo clearance appears under Docbot in the FDA device databases. On November 21, 2022, Satisfai Health, Inc. of Vancouver announced it had bought Docbot’s intellectual property and assets and transferred key technical staff. Terms were not disclosed. Satisfai now operates as Dova Health Intelligence, offering ulcerative colitis, Barrett’s esophagus and inflammatory bowel disease products for research and clinical trials, with a medical device in development.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| FDA status | No 510(k) and no De Novo grant under Docbot, Inc. as of Sep 23, 2026. Every company release labeled UltivisionAI an investigational device not cleared by the FDA. | Public record |
| AIDA, as registered | NCT04196088. Randomized, 978 estimated enrollment, primary endpoint mean adenomas per procedure. Started Jun 11, 2020. Last update posted Mar 19, 2021. Status unknown, last known recruiting. No results posted. | Public record |
| AIDA, as announced | The June 2020 release says 978 subjects “across 6 sites in the U.S.” The registry lists nine facilities in seven states. | Differs from registry |
| Performance claims | The CEO told BioWorld the system had shown 99 percent sensitivity and 98 percent specificity. An investor wrote that a preliminary single center study improved detection rate by 37 percent. Neither figure appears in a published randomized trial. | Company-stated |
| Peer reviewed papers | Five PubMed records list Docbot authors: polyp pathology prediction (2020), ulcerative colitis central reading with Eli Lilly (2021), Barrett’s neoplasia detection (2020), bowel preparation assessment (2022) and a 2020 AI summit report. | Public record |
| Federal funding | NSF SBIR Phase I award 1843975, $225,000, Feb 1 to Sep 30, 2019. No NIH award to Docbot, Inc. found. | Public record |
| Founding date | The 2018 release says the company was founded in 2014. The company website says the founders teamed up in 2013 and launched the company in 2016. | Company sources disagree |
| Corporate identity | Docbot is not Iterative Scopes and not Iterative Health. That is a separate Cambridge, Massachusetts company whose SKOUT device holds 510(k) K213686, cleared Aug 12, 2022, plus five later clearances through Dec 22, 2025. No record connects the two. | Public record |
| Patents | The acquirer’s release refers to Docbot’s patented technologies. No patent numbers appear in the records pulled here. | Not found |
Read plainly: Docbot produced better science than most AI startups its size, in the right journals, with a pharmaceutical partner and a federal grant. What it never produced was the thing that turns a paper into a product. The trial meant to carry UltivisionAI to the FDA went quiet eight months after it opened, and the company was sold for parts eighteen months after its Series A.
What to watch
- Whether AIDA data is ever published. The trial randomized patients and has never posted a result.
- Whether Dova Health Intelligence files for FDA clearance on anything built from the Docbot stack. It holds none as of September 2026.
- The wider CADe evidence base, where randomized gains have not reproduced in practice and a deskilling signal is now documented.
In their words
“If everyone of screening age got a colonoscopy and every physician was doing the procedure well, we could reduce the risk by 90 percent nationally”
William Karnes, MD, co-founder, UC Irvine News, 2019 · Independent
“While you’re scoping, if a polyp is detected, it will put it in a green box on the screen. It helps physicians in identifying polyps, especially hard to find polyps like the small polyps and the flat lesions.”
Andrew J. Ritter, CEO, to BioWorld, 2021 · Independent
“In the future, if you are getting a procedure like a colonoscopy done, your colonoscopist will be aided by an AI system ensuring you get the best result possible. You won’t want to have one done without it.”
Alex Morgan, MD, PhD, Partner, Khosla Ventures, Series A release, 2021 · Company release
“We believe our study is the first to demonstrate that artificial neural networks can be trained to predict levels of ulcerative colitis severity from routinely obtained full-length endoscopy videos, rather than curated still images, with high inter-rater agreement to human central readers.”
James Requa, Head of AI, Docbot, October 2020 release · Company release
“I am personally very grateful to the Docbot CEO, Andrew Ritter, and the founding physicians, Dr. William Karnes and Dr. Jason Samarasena, for trusting us to further develop and commercialize their patented technologies after so many years of dedication and hard work.”
Dr. Michael Byrne, CEO and founder, Satisfai Health, acquisition release, 2022 · Acquirer release
“Continuous exposure to AI might reduce the ADR of standard non-AI assisted colonoscopy, suggesting a negative effect on endoscopist behaviour.”
The Lancet Gastroenterology and Hepatology, 2025 · Independent
Related companies
Sources
- Public recordFDA device databases, no clearance under Docbot, Inc.
- Public recordClinicalTrials.gov NCT04196088, AIDA
- Public recordNSF SBIR Phase I award 1843975
- Public recordCentral Reading of Ulcerative Colitis Clinical Trial Videos Using Neural Networks
- Public recordFDA 510(k) K213686 and five later SKOUT clearances
- Public recordSEC EDGAR, no Form D filed by Docbot, Inc.
- IndependentBeating colon cancer with AI
- IndependentDocbot Exits Y Combinator, Receives $2.045M in Funding
- IndependentDocbot lands a healthy $4M in series A financing
- IndependentLack of Effectiveness of Computer Aided Detection for Colorectal Neoplasia
- IndependentEndoscopist deskilling risk after exposure to artificial intelligence
- InvestorWhy we invested in Docbot
- CompanyDocbot Closes Series A Financing Led by Khosla Ventures
- CompanyFirst Subjects Enrolled in Docbot’s Pivotal Study
- CompanyDocbot about and team page
- AcquirerSatisfai Health Purchases Assets of Docbot, Inc.
- AcquirerDova Health Intelligence, previously Satisfai Health
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
