Cerebriu

The models are the product. Apollo reads brain MRI while the scan is running. Apollo Acquire flags infarcts, intracranial hemorrhage and tumors from the first images and recommends which sequences to add before the patient leaves. Apollo Interpret ranks the finished studies by urgency and highlights key images. Every function is a model reading images; there is no scanner, archive or service layer underneath.

The routes to market make the same point. Cerebriu sells Apollo embedded in Siemens Healthineers MRI systems and through AI platforms and distributors, placing its model inside infrastructure others own. Remove the models and nothing is left to sell.

The Apollo page promises completeness. It says Apollo Acquire adapts the protocol "so every study leaves the scanner complete, with the sequences the finding requires already acquired." That is a promise about the product's output with no exception, and Cerebriu's own figures contradict it. Its published protocoling results give first stage sensitivity of 62.5 percent for tumors and 76.6 percent for hemorrhage, rising to 75.5 and 84.4 percent after refinement. So some studies with those findings will leave without the sequences they need. That is the failure that matters here, because a sequence not acquired cannot be read later and the remedy is recalling the patient.

The oversight itself is described in part. The scan starts as normal, Apollo recommends a protocol change that staff accept in one click, and the page says Apollo does not replace clinical judgment and the physician keeps final responsibility for diagnosis. Neither control reaches the miss: the one click checks a change Apollo suggests, and the physician's responsibility comes after the scan, so nothing surfaces a finding Apollo did not flag while the patient is still on the table. Nothing says what staff see when the model is unsure, or whether declined suggestions are recorded.

The approach and the version are named. The product is Cerebriu Apollo 2.1.2, with Smart Priority, Smart Protocol and Smart Reading as its features, sold as Apollo Acquire (during the scan) and Apollo Interpret (after it). Cerebriu's protocoling work describes a two stage model. A first stage reads the opening sequences and a second stage refines the result, with performance reported for each of the three target findings. Its publications list covers stroke lesion segmentation, DWI-FLAIR mismatch and selective gadolinium use.

The company publishes nothing on the model architecture, the size and source of the training data, or how updates are released and announced.

The model layer is Cerebriu's own. Apollo 2.1.2 is a CE marked device, so its model cannot be swapped without regulatory change, and no outside model provider is named in the path. In the Siemens Healthineers route the software runs inside the manufacturer's scanner.

The other parties are not named. The site lists none of the AI platforms and distributors that resell Apollo, and no hosting provider or subprocessor list is published for any route. Training data provenance is not described.

Published research stands behind the product, short of independent outcome evidence. Cerebriu's site lists peer reviewed work on MRI stroke detection, DWI-FLAIR mismatch and AI guided selective use of gadolinium contrast (Communications Medicine). It lists conference results for its protocoling model: first stage infarct sensitivity 90.1 percent and specificity 92.8 percent, with lower tumor and hemorrhage sensitivity. A 2023 European Journal of Radiology study of 800 scans reporting about 89 percent sensitivity and 90 percent specificity for ischemic lesions is also attributed to the product. A case study describes a safe reporting backlog.

No study shows that adapting protocols during the scan reduces recalls, speeds treatment or finds more disease in practice, which is what the product is sold on. Case counts are missing from the protocoling results on the site.

Nothing published describes how patient images move through Apollo. The only privacy document on the site is a website privacy policy, last updated June 2021, covering visitors, cookies and marketing. The Apollo pages say nothing about where analysis runs, what is stored, for how long, or whether customer scans are used to improve the models. No guardrails or safety event process is published.

The routes to market make the question practical. Apollo is sold embedded in Siemens Healthineers scanners and through AI platforms and distributors, and each route can move images differently.

No status is stated and no privacy document covers the service. The site's only privacy policy governs website visitors and was last updated in June 2021. Nothing addresses HIPAA, a business associate agreement, or, for European customers, Cerebriu's role as a GDPR processor and its data processing agreement.

Apollo is marked for investigation use only in the United States, so it is not sold clinically there today. That changes if FDA clearance comes.

No security certification, attestation, security page or trust center is published. The site footer holds a website privacy policy, a code of ethics and a support portal login, and nothing else on security.

Apollo ships inside Siemens Healthineers MRI systems, so it will have passed that manufacturer's own supplier and product security review. That review is not published and is not an outside audit a hospital can rely on.

The regulatory position is European, stated plainly on every page. Cerebriu Apollo 2.1.2 is regulated as a medical device in the EU. It conforms to the Medical Device Directive 93/42/EEC and is covered by Article 120 of the MDR, the transition provision for devices certified under the older directive. The homepage gives the class as IIa. In the United States it is for investigation use only, with no FDA clearance.

Cerebriu also sells Smart Protocol embedded in Siemens Healthineers MRI systems, which the company says was CE marked in 2025 as the first clinical AI embedded in a major manufacturer's scanner.

Cerebriu publishes a Code of Ethics in its footer and reports performance separately for each finding the product looks for. Its protocoling model's first stage sensitivity was 90.1 percent for infarcts, 76.6 percent for hemorrhage and 62.5 percent for tumors. That shows where the product is weaker by condition, but gives no counts and no results by patient group.

No results are published by age, sex, scanner maker, field strength or site. The failure mode here is quiet: if the protocol model under recommends a sequence for one group, that group's studies are incomplete, and later review of the scan will not show what was never acquired.

Error rates by finding are published for the protocoling model, without the counts or method needed to hold the company to them. First stage sensitivity was 90.1 percent for infarct (specificity 92.8 percent), 76.6 percent for hemorrhage and 62.5 percent for tumor. After a second stage it was 84.4 percent for hemorrhage and 75.5 percent for tumor, with specificities above 90 percent. The page gives no number of cases and no study design. A 2023 European Journal of Radiology study of 800 scans is attributed to the product; whether Cerebriu led or published it is not stated.

No warranty, indemnity or remedy is attached to a missed finding or a missing sequence, and the miss here is costly: the patient has to be recalled.

One named system is integrated deeply. Smart Protocol runs inside Siemens Healthineers MRI systems, acting at the scanner console during the scan. Cerebriu also sells Apollo through AI platforms and distributors, and Apollo Interpret ranks the worklist by urgency across scanners and sites.

No PACS, RIS or record system is named. No integration standard is described. Whether capability differs between the Siemens embedded version, the platform versions and a standalone installation is not stated.

The routes are stated; where the software runs is not. Apollo reaches hospitals embedded in Siemens Healthineers MRI systems, or through AI platforms and distributors. Acting during the scan implies analysis fast enough to run close to the scanner, but no page says so.

Nothing is published on hosting, region or retention for any route. A hospital buying through a platform may inherit that platform's hosting rather than Cerebriu's.

No price is published, but the way Apollo is sold is visible from the site's navigation. It reaches hospitals three ways: embedded in Siemens Healthineers MRI systems, through AI platforms and distributors, and directly through a demo request. Apollo Acquire and Apollo Interpret are presented as separate products. Where Apollo comes inside a manufacturer's scanner or a platform subscription, the partner may set the price.

Coverage is named precisely. Apollo covers brain MRI only, for three findings: acute and subacute infarct, intracranial hemorrhage and intracranial tumor. It is sold to four settings: hospital radiology departments, private radiology centers, stroke centers and teleradiology providers. The homepage says it is deployed across Europe and the Middle East, and it is for investigation use only in the United States.

Validation stands behind part of that. Performance is published for each finding, with infarct detection strongest and tumor and hemorrhage detection weaker. There is nothing for other body areas, other modalities or children.