VUNO

The deep learning model is the product. DeepCARS converts four routinely charted vital signs into a 0 to 100 risk of cardiac arrest in the general ward within 24 hours, and there is no hardware, monitoring device or services layer sold alongside it. The company describes it as Korea's first AI medical device for this purpose, and as a biosignal product in a home market where medical AI had been largely image based.

The company also sells imaging AI, VUNO Med-Chest X-ray and VUNO Med-Fundus AI, where the model is likewise the product.

Produces a risk score for a rapid response team to act on, so a person decides whether and how to intervene. The score runs from 0 to 100 for the risk of cardiac arrest in the general ward within 24 hours, and the company positions the tool as enabling timely intervention rather than triggering automated escalation; nursing staff, not the algorithm, mobilize the response.

The oversight question specific to early warning systems is the threshold. A score that fires too often produces alarm fatigue and gets ignored, while one tuned too conservatively misses patients. No recommended threshold or guidance on setting one is published, which is the missing part.

Inputs and outputs are stated precisely, four named vital signs producing a 0 to 100 score over a 24 hour horizon, and the approach is named as deep learning on vital sign sequences. Discrimination is published separately for internal and external validation, 0.860 and 0.905 in a multicenter study of 173,368 patients, and reporting the external figure on its own shows the model was tested outside its development setting.

Architecture detail, training data composition and version history are not published, and no calibration is reported, which matters for an early warning system because the score drives a threshold and the threshold drives alerts.

A structurally short chain with its processors named. The models are the company's own, with DeepCARS approved in Korea, so no outside model provider sits between the record and the score. The privacy policy names the processors it entrusts personal information to, including AWS for cloud infrastructure, Google for analytics, Naver Cloud for messaging and CSRcube for clinical trial data capture, and states retention periods by product and that nothing is transferred overseas.

What keeps this from A is which party handles which product's data, and the terms on development use: the policy says pseudonymized information is used for AI software development without saying whether it comes from data processed for hospitals.

Peer reviewed, multicenter and prospective. The multicenter validation of 173,368 patients, published in Resuscitation in 2021 and cited on the DeepCARS page, reports an area under the curve of 0.860 internally and 0.905 externally. A prospective multicenter validation published in Critical Care in 2023, with Seoul National University Hospital, Seoul National University Bundang, Inha and Dong-A University Hospital, tested prediction of cardiac arrest or unplanned ICU transfer and reported better performance than conventional early warning scores. Further papers appear in the Journal of the American Heart Association and Critical Care Medicine.

Prospective evidence is uncommon for deterioration prediction tools, most of which rest on retrospective cohorts. All published evidence comes from Korean cohorts, so performance elsewhere is untested.

Retention is stated product by product, and the use of data for AI development is stated. The privacy policy (effective 23 March 2026) says the company processes patient information on behalf of healthcare institutions for its AI medical device services, including name, sex, date of birth, patient identification number and clinical information, and gives retention periods: DeepCARS data until the service ends, and Chest X-ray and Fundus AI images for five years from storage. It states that pseudonymized information is processed for clinical trials, AI software development and related research, and that personal information is not transferred overseas.

What keeps this from A is the source of the development data. The policy does not say whether information processed for a hospital is pseudonymized and used to develop the company's models, or whether a hospital can decline that use. Safety engineering, meaning guardrails and how a safety event is handled, is not published. The policy covers the Korean service; the US subsidiary's handling is not described there.

An equivalent regime, documented to depth for the Korean service. The privacy policy (effective 23 March 2026) states that the company processes patient information on behalf of healthcare institutions, the processor role under Korean law that corresponds to a business associate, and backs it with product level retention periods, named processors, a statement that personal information is not transferred overseas, and ISO/IEC 27701 privacy certification.

No HIPAA position is published. The company holds two US clearances and sells through a US subsidiary, VUNO MED Inc., and whether it signs business associate agreements, which entity signs, and whether US patient data is processed outside the United States are not stated.

Two recognized certifications are named with version, scope and validity. The company's site states ISO/IEC 27001:2022 and ISO/IEC 27701:2019, both valid from 24 November 2025 to 23 November 2028, for the provision of medical services including the development and supply of solutions and SaaS, and the privacy policy repeats them.

No trust center or retrievable certificate is published, and no penetration testing or vulnerability disclosure statement.

Status is split across the product line and by market. DeepCARS, the lead product, is approved by Korea's Ministry of Food and Drug Safety and carries a CE mark, and in the United States holds FDA Breakthrough Device designation; no 510(k) for it appears in the FDA database. A designation is a commitment to a review pathway, not an authorization, so a US hospital cannot deploy it clinically on that basis.

The imaging line has US clearances: VUNO Med-DeepBrain, brain MRI quantification, as K231398 (October 2023), and VUNO Med-Chest X-ray Triage as K241439 (November 2024), both filed by Vuno, Inc. A clearance for one product does not extend to another.

Multicenter validation across several independent Korean hospitals addresses site level generalization, and the prospective study tests the model in live conditions rather than on curated retrospective data. What is absent is any subgroup analysis by age, sex or comorbidity, and the concern is concrete for this product class: vital sign based deterioration models are known to perform differently across those groups.

The entire published evidence base is drawn from one national population, which leaves performance in more heterogeneous populations, including the United States, undocumented.

The score is advisory and a person acts on it: DeepCARS gives a rapid response team or ward staff a risk score, and they decide whether to assess or escalate. That review before action is the mechanism that lets a wrong score be caught.

What is published about accuracy is discrimination, not an error rate. The DeepCARS page cites an area under the curve of 0.860 on internal validation and 0.905 on external validation in a multicenter study of 173,368 patients (Resuscitation, 2021), and an average of 15.8 hours between the first alert and the arrest. An area under the curve ranks patients against each other; it does not say how often an alert is false or an arrest is missed at the threshold a hospital uses, and alert rate is what decides whether nurses keep responding. No sensitivity, specificity or alerts per patient day at an operating point, no calibration and no warranty, indemnity or remediation commitment is published.

Integration is implied by the product and not documented. DeepCARS takes four routinely charted vital signs, which live in the hospital record, and returns scores into ward workflows, and the company frames the reliance on routine vital signs as an adoption advantage because no new instrumentation or documentation is required.

No record system, interface standard, connector or API is named, and integration with Korean hospital record systems does not carry over automatically to Epic or Oracle Health environments.

Hosted, with residency stated for the Korean service. The company's ISO/IEC 27001 certificate scope covers the development and supply of solutions and SaaS, its privacy policy names AWS as cloud infrastructure among its processors, and the policy states that personal information is not transferred overseas.

What is left open is isolation and the path for other markets. Whether hospitals can run the products on premise, how customers are separated in the hosted service, and where data from US or European customers is processed are not published.

No price is published, and how DeepCARS is paid for in Korea is. The company's history page records that in 2022 DeepCARS became the first AI medical device to enter Korea's non-reimbursed market, and the first designated under the country's early access medical technology route: hospitals can charge patients for it as a non-covered service while its formal evaluation continues. That tells a Korean hospital how the cost is recovered, not what the vendor charges. The company has been listed on KOSDAQ since 2021, which brings financial disclosure and some visibility into vendor stability.

Nothing is published on per bed, per hospital or subscription cost. In the United States, DeepCARS holds Breakthrough Device designation, which is not an authorization, so US terms are an open question.

Narrow for the lead product, and validated in it. DeepCARS covers admitted inpatients on general wards, predicting cardiac arrest within 24 hours, where rapid response teams are meant to intervene; it does not extend to intensive care, emergency triage or outpatient monitoring. Retrospective and prospective multicenter studies in Korean hospitals validate that setting; performance outside Korea and by hospital type is not published.

VUNO appears among imaging AI vendors for a second line it sells to radiology and eye care buyers: VUNO Med-Chest X-ray, cleared in the United States for triage, and VUNO Med-Fundus AI. It also holds a US clearance for VUNO Med-DeepBrain, brain MRI quantification.