icometrix

The measurements are the product. icobrain automatically labels, segments and measures brain structures and lesions on MRI and CT. It reports volumes and changes over time against an age and sex matched reference, and icobrain aria detects, measures and grades amyloid related imaging abnormalities. Every module sold is one of these readings, organized by condition.

The surrounding software carries no value of its own. icobridge moves images from the hospital to the analysis and returns results, and the icompanion app gives patients with multiple sclerosis a view of their scans. Neither would be bought without the measurements.

The homepage FAQ promises icobrain delivers "consistent, reproducible brain measurements regardless of scanner type or protocol." That is a reliability promise about the output with no exception, and icometrix's own FDA filing narrows it. The K192130 summary says the longitudinal pipeline measures change only between two MR scans from the same scanner, with the same protocol and contrast, and that single time point and longitudinal results cannot be compared. Volume measurements shift with scanner and protocol. Anyone who takes the FAQ at its word may read a scanner change as atrophy.

The oversight design is otherwise sound. Color coded segmentations come back in DICOM so the reader sees what was measured, and results arrive as prepopulated reporting templates the radiologist edits. The FAQ says icobrain supports decisions rather than making them. The smallest change the system can reliably detect is not published.

The pipeline and the version are named. The K192130 summary describes icobrain 5.0 as two pipelines, icobrain cross for single time point volumes on MR or CT and icobrain long for change between two MR scans. Each converts DICOM, segments, measures and produces a report with overlays. It lists the standards the software was built under (ISO 14971, IEC 62304, IEC 62366, ISO 13485). Accuracy and reproducibility were tested against simulated and manually labeled ground truth and test retest scans. The outputs are named measure by measure.

Left unpublished are the model architecture, the training data, the makeup of the normative reference population, and how updates are released.

The chain is structurally short. The models are icometrix's own and FDA cleared, so none can be swapped without new review, and no outside model provider sits in the path. Only pseudonymized, encrypted studies leave the hospital, through the icometrix owned icobridge gateway.

The host behind the icometrix cloud is not named and no subprocessor list is published. Training data and the normative reference population are not sourced. icometrix is now a GE HealthCare company, and whether hosting or data flows move to the parent is not stated.

icometrix publishes a large body of research and headline results from it, short of independent outcome evidence. The site cites more than 250 scientific articles. Its homepage figures come from company authored studies (Van Hecke et al., 2021; Sima et al., 2021): 150 percent more multiple sclerosis disease activity detected, 40 percent faster reading, 35 percent more consistent reading, and treatment failure detected 2.5 times faster. The UK's NICE issued a Medtech Innovation Briefing on the multiple sclerosis measures in 2022. Named sites include VITAZ, Rush, Shields Healthcare and DENT Neurologic Institute.

The homepage also claims a "68% outcome improvement when employing icobrain vs drugs alone" and gives no source for it. No study shows that using the measurements changes patient outcomes.

A specific commitment is published on the clinical workflow page. Scans leave the hospital only after the icobridge gateway inside the hospital network has pseudonymized and encrypted them. The icometrix cloud analyses pseudonymized data, and results are decrypted and re-linked inside the hospital before they reach PACS or the record system. That is checkable, and it keeps identified images inside the hospital.

What is not published: how long the cloud keeps pseudonymized studies and prior measurements, which comparison over time requires, and whether customer scans are used to develop or retrain models. No guardrails or safety event process is published.

Business associate status is stated and the agreement is offered. The clinical workflow page lists a Business Associate Agreement template and a Data Processing Agreement template among the documents a hospital can request, alongside a Terms and conditions page and Privacy Policy in the footer. The same page describes how patient data moves: the icobridge gateway in the hospital pseudonymizes and encrypts each study before it goes to the icometrix cloud, and decrypts results on the way back.

The agreements are not published, so their terms on subcontractors, retention and breach notice cannot be read before contracting. icometrix is now a GE HealthCare company.

icometrix names ISO 27001 among the certificates it provides on request (with ISO 13485, MDSAP, FDA and CE). The same page offers pre filled security questionnaires: the MDS2 medical device security form and the Cloud Security Alliance CAIQ. The footer links a vulnerability reporting page.

No artifact is published: the ISO 27001 certificate, its scope and expiry are behind a request, and no SOC 2 report or penetration testing summary is mentioned.

icometrix holds six FDA 510(k) clearances. Four are for icobrain brain quantification (K161148, August 2016; K180326; K181939; K192130, December 2019, icobrain 5.0 on MR and non contrast CT). One is for icobrain-ctp (K192962, February 2020). icobrain aria (K240712, November 2024) detects, measures and grades amyloid related imaging abnormalities under product code QBS, which the company calls the first AI computer aided detection and diagnosis clearance in neuroradiology. It is CE marked through notified body 1639, and the workflow page lists MDSAP and ISO 13485 certificates on request. The cleared longitudinal pipeline measures change only between MR scans from the same scanner, protocol and contrast (K192130).

The homepage counts nine FDA cleared solutions against six clearance numbers, so some condition modules sit under the general icobrain clearance.

icobrain has received the American College of Radiology's Transparent AI badge, which icometrix announces on its site. The ACR program reviews the disclosure material a vendor provides for a customer's own evaluation. Reports compare each patient with an age and sex matched reference population, which builds the main demographic effect on brain volume into the output.

No performance by patient group is published: nothing by sex, age band, race or ethnicity, scanner maker or field strength. The make up of the normative reference population, which every comparison relies on, is not described.

A reader can challenge each output. Results come back as color coded segmentations over the original images and as prepopulated report templates the radiologist edits, so the basis for every number is on screen before it enters a report.

No error rate is published in a form a buyer can hold icometrix to. The K192130 summary reports agreement averaged across all experiments: Pearson 0.91 and intraclass correlation 0.90 on 463 MR datasets, and 0.94 and 0.93 on 618 CT datasets. It gives no figure per structure or indication. No warranty, indemnity or remedy is attached. The icobrain aria summary (K240712), a detection device, may carry sensitivity and specificity with counts.

Integration standards and output formats are published with named deployments. The icobridge gateway takes DICOM from the scanner or PACS and returns results as DICOM secondary capture, encapsulated PDF or structured report, an HL7 PDF report, and prepopulated reporting templates, into PACS, RIS or the record system. The workflow page says icobridge installs in five minutes, with a default turnaround of ten minutes. Named sites include VITAZ, Rush and Shields Healthcare, and a platforms section shows logos of integration partners.

No record system or PACS product is named in text, and there is no marketplace listing or documented bidirectional record integration.

One hosted model is described. The icobridge gateway sits inside the hospital network, pseudonymizes and encrypts each study, and sends it to the icometrix cloud for analysis. Results come back to the gateway for decryption and delivery to PACS or the record system. The workflow page claims zero percent downtime.

Where the icometrix cloud runs is not stated: no provider, region or country, and no statement on whether European and US data are processed separately. No on premises option is described. How long pseudonymized studies stay in the cloud for comparison over time is not stated either.

No price is published. The homepage and reimbursement page describe the payment route instead. Two Category III CPT codes, 0865T and 0866T, are in use across US regions, and icometrix says more than 30 insurers have processed claims under them. That is what a provider may recover, not what icometrix charges.

The product is sold by condition (multiple sclerosis, dementia, amyloid monitoring, stroke, trauma, epilepsy, Parkinson's, tumors), as a cloud service reached through the icobridge gateway. icometrix is now a GE HealthCare company, which may change how it is sold.

Coverage is named across eight neurological conditions: multiple sclerosis, dementia and Alzheimer's disease, amyloid related imaging abnormalities, stroke, traumatic brain injury, epilepsy, Parkinson's disease and brain tumors, on MRI and CT. icometrix says it is used in more than 400 hospitals in over 30 countries.

Validation stands behind part of it. FDA lists six icometrix clearances. Four cover general icobrain quantification of brain structures on MRI and CT, tested on healthy people and patients with Alzheimer's, multiple sclerosis, brain injury and depression. One covers CT perfusion and one icobrain aria. The homepage counts nine FDA cleared solutions, but the condition modules do not each have their own clearance. The core filing also limits longitudinal change to scans from the same scanner and protocol.