Brainomix

The models are the product. Brainomix 360 Stroke reads non contrast CT, CT angiography, CT perfusion and MRI to score early ischemic change (ASPECTS), estimate core volume and net water uptake, and flag suspected large vessel occlusion and hemorrhage. Brainomix 360 e-Lung quantifies features on chest CT. Each is a cleared device whose output is the model's reading.

The rest of the offer depends on that output. Vantage is an analytics dashboard built on the AI results across a stroke network, and the life sciences work sells the same imaging biomarkers as trial endpoints. Remove the models and nothing is left to sell.

The oversight structure is published in the FDA summary for Brainomix 360 Triage Stroke (K251983). The module runs in parallel with normal reading and sends a notification only. It is not a rule out device, and cases it does not flag should still go through angiography. Mobile images are compressed previews not meant for diagnosis. It is not for patients with recent neurosurgery, recent stroke or bilateral MCA occlusion, and it is validated on Siemens, GE and Philips scanners only. The FAQ says every module is decision support with final decisions left to the clinician. Clinicians can edit automated ASPECTS results, and edited results are labeled as such in every output.

The alert threshold behind each flag is not published.

Methods are named in part. The K251983 summary says Triage Stroke finds large vessel occlusion on non contrast CT by combining its ASPECTS and hyperdense vessel sign algorithms. It says the device uses machine learning, other artificial intelligence and non adaptive imaging algorithms. The FAQ describes a patented voxel wise heatmap behind the ASPECTS score. On the lung side the company has published a named biomarker, the weighted reticulovascular score. More than 80 peer reviewed papers describe the methods and validation.

Network architectures, training data size and sources, and the way updates to each module are released and announced all go undescribed.

The chain is structurally short. The models are Brainomix's own and each module is FDA cleared or CE marked, so none can be quietly swapped, and no outside model provider sits in the path. In the on premises deployment the software runs on a server inside the hospital and, the company says, patient data does not leave it.

The cloud configuration adds parties that are not named: no cloud provider, region or subprocessor list is published. Training data provenance is not described either, which matters for a company that also runs imaging biomarker work with pharmaceutical partners.

The stroke evidence is large, peer reviewed and multi site, and it measures treatment rather than accuracy alone. Brainomix reports a study in The Lancet Digital Health (2025) covering 452,000 patients at 107 hospitals over five years. It associates Brainomix 360 Stroke with a 100 percent relative increase in mechanical thrombectomy rates. The company lists more than 80 peer reviewed publications, among them a multicenter study in the Journal of NeuroInterventional Surgery (2024, 111 patients) finding core volume from non contrast CT equivalent to CT perfusion. It also cites a multicenter study finding its ASPECTS scoring non inferior to expert scorers.

The thrombectomy figure is an association from a staggered real world rollout, not a randomized comparison, so other pathway changes over five years can contribute. Lung evidence is thinner and sits largely in pharmaceutical trial work.

The Trust Center FAQ makes categorical commitments. Patient data is stored on the Brainomix server inside the hospital and is not transmitted externally. Email notifications carry pseudonymized results only. Results reached remotely go through the hospital's own VPN or Citrix, or the Brainomix mobile app. The mobile previews are compressed images not meant for diagnosis.

The company publishes no retention schedule and does not say whether customer imaging is used to train or improve models, which matters for a company that also sells imaging biomarkers to pharmaceutical partners. What the cloud hosted configuration changes is also unstated. No guardrails or safety event process is published.

A position is stated without the document behind it. The Trust Center FAQ says patient data is stored on the Brainomix server inside the hospital and is not transmitted externally, and that email notifications carry pseudonymized results only. It says Brainomix is ISO 27001 certified and GDPR compliant, with a Data Processing Agreement available on request. The privacy policy linked from the site is an EU policy.

Two gaps matter for a US buyer. Nothing mentions HIPAA or a business associate agreement, though the company sells FDA cleared products from a Chicago office. And the no transfer statement does not fit every configuration: the FDA summary for K251983 says analysis runs on the Brainomix server on premises or in the cloud, and the FAQ describes cloud hosted appliances.

A recognized certification is named in Brainomix's own material: the Trust Center FAQ says Brainomix is ISO 27001 certified, alongside GDPR compliance and a Data Processing Agreement on request.

No artifact is published. The certificate, its version, scope, issuing body and expiry are not shown. The Trust Center is a set of frequently asked questions rather than a portal holding documents. No SOC 2 report, penetration testing statement or vulnerability disclosure route is published.

Brainomix Limited holds 13 FDA 510(k) clearances, from Brainomix 360 e-CTA (K192692, May 2020) to Brainomix 360 Hyperdensity (K260406, June 2026). They cover e-ASPECTS (K221564, K243294), e-CTP (K223555), e-MRI (K231656), e-CTA (K242123), Triage ICH (K231195), Triage LVO (K231837), and Triage Stroke (K232496, K251983). They also cover e-Lung (K233875, K242411). The FAQ states CE marking as Class IIb under the EU MDR.

Clearance is per module. The collateral score and DWI-FLAIR mismatch are CE marked only. The triage modules are notification only and not for diagnosis. The homepage cites more than 35 regulatory clearances worldwide.

Brainomix's own FDA summary for Brainomix 360 Triage Stroke (K251983) publishes performance by group, with numbers. Results are given for age band, sex, CT slice thickness and scanner maker, for both hemorrhage and large vessel occlusion detection. That lets a buyer see where the product is weaker. For large vessel occlusion on non contrast CT (112 positives), sensitivity was 84.6 percent under age 50, 74.4 percent at 50 to 69 and 63.3 percent at 70 and over. It was 62.5 percent on Siemens scanners against 77.8 percent on Philips, and 61.4 percent on the thinnest slices. For hemorrhage the spread was narrower, but the Siemens group had only 7 positives. The summary says more subgroup results are in the labeling.

Race and ethnicity are not reported, and several subgroups are small with wide confidence intervals. Nothing like this is published for e-Lung.

Brainomix publishes error rates for its stroke triage software in its own FDA filing, with method and denominator. In the K251983 summary for Brainomix 360 Triage Stroke, three neuroradiologists set ground truth by consensus in retrospective studies. Hemorrhage detection on 341 cases (167 positive) had sensitivity 96.4 percent and specificity 96.6 percent. Large vessel occlusion on non contrast CT, across 267 cases with 112 positive, had sensitivity 69.6 percent (95% CI 60.7 to 77.7) and specificity 89.6 percent. A reader study found the device more sensitive for large vessel occlusion than general radiologists reading the same scans (47.2 percent).

The large vessel occlusion figure means about three in ten occlusions on non contrast CT go unflagged, which is why the filing says the device is not a rule out and every patient still needs angiography. No warranty, indemnity or remedy is attached.

Integration is DICOM based and described concretely. Brainomix 360 receives studies from scanners and PACS. The FAQ says it supports all major scanner and PACS makers, and Brainomix engineers test every connection before go live. Results return to PACS, and also go out by pseudonymized email, the mobile app and a web browser, typically within one to two minutes. Named deployments include Örebro in Sweden and hospitals in the English health service.

No EHR integration is named, and no HL7 or FHIR path is described, so whether a result reaches the patient record as a discrete entry is not stated.

Options are stated. The FAQ says Brainomix 360 is provided as a managed appliance on virtualised (cloud) or physical hardware. The K251983 summary says it runs on an off the shelf physical or virtual server, with analysis on the Brainomix server on premises or in the cloud. For the on premises case residency is stated: patient data is stored on the Brainomix server inside the hospital and not transmitted externally.

The cloud path is left open. The same FAQ answer says data does not leave the hospital, which a cloud hosted appliance cannot mean in the same way. No cloud provider or region is named, and isolation between customers is not described.

No price is published, but the way the product is sold is. The FAQ says each Brainomix 360 Stroke module (non contrast CT, CT angiography, CT perfusion, MRI) is licensed on its own or as a package. e-Lung and the Vantage dashboard are separate products, and installation is a managed appliance set up by Brainomix engineers. A buyer can see that cost depends on which modules and sites are covered, but not the unit or the figure.

Coverage is named: acute stroke across non contrast CT, CT angiography, CT perfusion and MRI, plus interstitial lung disease on chest CT. The company says it runs in more than 400 hospitals in over 20 countries, from regional hospitals to academic centers. The FDA summary for Brainomix 360 Triage Stroke (K251983) names the validated setting for that module. It covers adults in the acute setting within 24 hours of symptom onset, on Siemens, GE and Philips scanners, excluding recent neurosurgery, recent stroke and bilateral MCA occlusion.

Validation stands behind part of the coverage. Some stroke modules, the collateral score and DWI-FLAIR mismatch, are CE marked only and not cleared in the United States. The split between comprehensive and referring centers is not published.