How AI Assists Surgeons Today: Five Systems, Five Published Studies
Updated: Oct 15
Most articles about AI in surgery describe a future state.
This one describes five systems already installed in operating rooms, with the clinical data behind each one.
Every claim below traces back to an FDA record or a peer-reviewed journal, not a press release.
Seeing Blood Flow Without Dye
ActivSight, made by Activ Surgical, attaches between a standard laparoscope and camera system.
It cleared FDA review as a 510(k) device on April 8, 2021, for real-time fluorescence and near-infrared imaging of vessels, blood flow, and tissue perfusion during minimally invasive surgery.
Instead of injecting indocyanine green dye and waiting for it to circulate, the system uses laser speckle contrast imaging to build a live, dye-free perfusion map from reflected laser light.
A first-in-human study followed 67 patients through cholecystectomy, colorectal, and bariatric procedures between November 2020 and July 2021, split between robotic-assisted and laparoscopic approaches.
Surgeons detected perfusion successfully in 97.5% of robotic cases and 100% of laparoscopic cases, with no adverse events reported.
In practice, that means a surgeon checking blood supply to a bowel segment before an anastomosis gets a live color map instead of a visual estimate.
Source: Surgical Endoscopy
Coaching Technique Through Video Review
Theator records laparoscopic cases and uses computer vision to flag key steps, including whether a surgeon achieved the "critical view of safety" before cutting the cystic duct during gallbladder removal.
Two general surgery departments, one affiliated with McGill University and one at Tel Aviv Sourasky Medical Center, used the platform to track that single benchmark across 279 cholecystectomies performed between December 2022 and August 2023.
After the departments set a 75% adoption target, monthly compliance rose from 33.3% to 75.7%, and overall adoption climbed from 39.2% to 69.2% (P < 0.001).
Procedures with full compliance were also shorter and had fewer adverse events than those without it.
The mechanism here isn't autonomy. It's a surgeon getting an objective readout of their own habits, case after case, until the habit shifts.
Source: Annals of Surgery
Writing Reports From Video, Not Memory
Most operative reports are dictated hours or days after the procedure, reconstructed from memory under time pressure.
A Mayo Clinic-led study tested whether Theator's computer vision engine could generate a more accurate report directly from surgical video, using 158 robotic-assisted radical prostatectomies as the dataset.
Surgeon-written reports contained at least one discrepancy against the actual video in 53.2% of cases.
The study's authors disclosed that Theator supplied only the step-detection algorithm; the Mayo Clinic team ran the analysis and wrote the conclusions independently.
Oracle Health began integrating the same platform into its EHR system under a partnership announced June 24, 2026, so structured operative data flows directly into a patient's chart.
Predicting Outcomes From Checklist Compliance
Surgical Safety Technologies' OR Black Box records audio, video, and physiologic data from the operating room, then uses AI-based video review to score whether the team completed each step of the surgical safety checklist.
A retrospective study at UT Southwestern Medical Center analyzed 4,581 patients treated in Black Box-equipped rooms between August 2020 and September 2022.
Full checklist compliance was independently associated with lower mortality odds (OR 0.96, P < 0.05) and shorter hospital stays.
"Debriefing," the step most teams skip when the room is running behind, carried its own independent association with mortality, length of stay, and ICU admission.
The system doesn't tell a surgeon how to operate. It tells a hospital, in aggregate, which routine behaviors are quietly costing patients.
Source: Surgical Endoscopy
Reducing Positive Margins in Breast Cancer Surgery
Claire, an OCT imaging system with an FDA-cleared AI algorithm from Perimeter Medical Imaging AI, scans excised lumpectomy tissue and flags suspicious regions before the surgeon closes.
Its pivotal trial enrolled 613 patients across 11 sites between December 1, 2021, and September 23, 2024, randomizing 206 patients into the device arm.
Standard care alone left 35 of those patients with at least one unaddressed positive margin; adding Claire brought that down to 28, a 20% relative reduction (P = 0.005).
Margin-level sensitivity rose from 48.7% to 61.5% with Claire added, at some cost to specificity — more true catches, a few more extra shaves.
In practice, that's a handful of patients per hundred spared a second surgery, not a device that reads margins perfectly.
What Connects These Five Systems
None of them cut, suture, or make a clinical decision.
Each one watches, measures, or visualizes something a surgeon would otherwise estimate, then hands the result back in real time.
That's the actual state of AI in the OR: augmentation with a paper trail, not autonomy.
Every claim about "AI-powered" surgical technology deserves the same scrutiny applied here: an FDA record or a named journal, not a vendor's own language.
The AI Claims Audit applies that standard across the surgical AI market's most repeated claims.
