MosaicOS
MosaicOS is a cloud radiology platform from Mosaic Clinical Technologies, the technology subsidiary Radiology Partners launched in July 2025. Radiologists built it for their own national practice first, and in June 2026 it went on sale to outside hospitals and health systems.
Mosaic Reporting is the product on sale today. It works alongside the radiologist during the read, following natural dictation and building a structured report in real time, with voice directed edits and guideline based follow up recommendations the radiologist confirms or changes before signing. By November 2026 it was turning out more than 100,000 reports a day for over 2,200 radiologists across 74 Radiology Partners practices, including Columbus Radiology. It runs beside an existing PACS, and a viewer, worklist, archive and information hub are on the roadmap.
Behind it sits Cognita, Mosaic's AI unit, which builds vision language models that read the images and prior studies and draft findings for the radiologist. Those drafting models are investigational and not for sale in the US. The chest X ray model received FDA Breakthrough Device designation in March 2026, and an internal CT study covered 71,707 exams. Radiology Partners also works with Stanford's AIDE Lab on monitoring AI in practice. Pricing is quoted for each organization.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
Reporting is the core of what Mosaic sells, and it is AI from end to end. Mosaic Reporting uses ambient AI and language models to follow a radiologist's dictation during the read and assemble a structured report as the radiologist speaks, so the report is built during interpretation rather than typed up after it. The investigational drafting layer goes further, with Cognita's vision language models reading the images and proposing findings.
Around those sits a platform with value of its own, running beside the PACS today and planned to absorb the viewer, worklist and archive. Take the models away and the reporting product would have little left to offer.
The radiologist signs every report. Mosaic Reporting builds the report from dictation, and the radiologist edits it by voice and confirms or changes the guideline based follow up recommendation before signing. In the research deployment, drafted findings arrive in the report for the reader to accept, edit or reject.
Mosaic monitors how often readers keep drafted content and how long they take to sign, watching for overreliance. What level of agreement triggers an intervention, what happens after a missed finding and when a model would be pulled are not stated, and monitoring cannot stop a wrong draft from being signed.
Two kinds of model run the platform. Language models turn dictation into a structured report in Mosaic Reporting. Cognita's proprietary vision language models, such as Cognita CXR for chest X rays, take the current exam, prior imaging and clinical context and write proposed findings in plain language, which the reporting layer then folds into the report. That separates image reading from report construction in a way a radiology buyer can follow.
Model version numbers, training populations and how customers hear about a model change are not given, and continuous platform updates leave open how one algorithm version is held steady during an evaluation.
Cognita, Mosaic's AI unit, builds the imaging models in house, inside a corporate chain that runs from Mosaic Clinical Technologies to its parent, Radiology Partners, rather than behind an unnamed outside API. That covers the investigational drafting models.
The suppliers behind the language models in the commercial reporting product, the cloud infrastructure and any third party diagnostic tools plugged into the platform are not named, and no subprocessor list sets out their retention, training or transfer terms. MosaicOS is built to host outside AI tools beside its own, so the chain grows with every one a site adds.
Mosaic's internal CT study is the strongest evidence, and it covers the investigational drafting models. It looked back at 71,707 exams, sent 3,366 selected cases to blinded expert review and assessed 696 reports for usability, with denominators and comparison methods stated. It is Mosaic's own research rather than independent validation.
For the reporting product, the evidence is scale: by November 2026 more than 2,200 radiologists across 74 affiliated practices were producing over 100,000 reports a day. Neither shows improved patient outcomes, and none of it comes from a health system outside Radiology Partners.
Images, clinical context and dictated findings all flow into a draft the radiologist reviews. Mosaic watches how readers interact with drafts, which gives it a way to spot overreliance and quality drift. The privacy framework separates patient health information from website data and points patients to Radiology Partners' clinical privacy notice.
How long source images and dictation are kept for the software, whether they train models and where deidentification happens are not stated. The monitoring has no published intervention thresholds or model incident procedure, and with the reporting product now sold beyond Radiology Partners, outside hospitals' studies join the same flow.
Mosaic's privacy policy sends questions about protected health information to Radiology Partners' Notice of Privacy Practices, and the site links to the radpartners.com policy rather than one of its own. That notice names the clinical organization, explains permitted uses and disclosures, and gives patients access, correction and complaint routes, so it reaches patient care rather than web analytics.
Its scope is the affiliated practices and their patients. For a hospital buying MosaicOS as software, Mosaic's role as a business associate, the agreement it signs and its subcontractor terms are not set out.
Resilience is described in some detail: redundancy, ransomware resistance and a 99.99% availability objective. Those are design goals rather than an audited security program. MosaicOS carries no SOC 2, HITRUST or ISO 27001 attestation of its own, and no trust page names a certified scope or audit period.
Clinical validation of the models and the size of the parent practice answer different questions from how patient data and the hosted infrastructure are protected. With outside hospitals now sending studies to the platform, an independent check of those controls is the missing piece.
The line between what is sold and what is in research is drawn plainly. Mosaic Reporting is commercially available. Cognita's drafting models are investigational, run under a research protocol aimed at FDA authorization and are not for sale in the United States.
The comprehensive chest X ray model received Breakthrough Device designation in March 2026, which speeds work with the agency but does not authorize sale. Some platform marketing still speaks of diagnostic benefits that depend on those research models, which runs a little ahead of what a hospital can deploy today.
Mosaic sets out a lifecycle for clinical AI: validation before deployment, user education, a controlled rollout, ongoing monitoring, and intervention followed by reassessment. For the investigational CT drafting, a monitoring dashboard compares model findings with signed reports to catch misses and overreliance. Radiology Partners also has a partnership with Stanford's AIDE Lab on methods for evaluating and monitoring AI in practice.
What is missing is a comparison by patient group. The CT test set included pediatric exams, but results for children against adults, or by sex or ethnicity, are not reported.
A radiologist can reject or rewrite any proposed finding before it becomes a signed report, and Mosaic tracks whether readers keep or change drafted content. The CT research gives model error figures with denominators, but for an investigational component rather than as a promise about the commercial reporting product.
No output warranty, indemnity or correction deadline comes with the software. Patients' rights to correct their records run to the treating practice under its clinical notice, not to Mosaic as the supplier of software to another organization.
MosaicOS is built to sit beside an existing PACS and take over more of the imaging workflow over time. Inside it, Mosaic Reporting takes dictation, builds structured reports and inserts approved follow up recommendations without the radiologist switching windows, replacing the older reporting system rather than layering on top of it.
What leaves the platform and where it lands is less clear. No EHR is named, no interface standards such as HL7 or FHIR are listed, and no deployment shows reports returning to a hospital's chart.
The platform is hosted, cloud native and updated continuously, and a hospital can run it beside its current PACS rather than replacing everything at once. Mosaic's stated 99.99% availability objective applies to that hosted service.
The cloud provider, storage region, tenant separation and backup location are not named, and neither is any cross border processing by support teams or model services. Radiology Partners operates in the United States, but that is not a residency commitment for the software, and no customer run deployment or region choice is offered.
Hospitals, health systems and radiology groups buy through a sales conversation that ends in an implementation plan sized to their requirements, system complexity and number of sites. The platform is modular, with more applications added as they are built and validated.
No subscription rate, per report or per user charge, license minimum, implementation fee or contract length is listed, and a demo is the way in rather than a trial or self serve plan. The drafting models cannot be bought at all while they remain investigational.
Radiologists in hospital and health system settings are the users, with Columbus Radiology among the named practices and 74 Radiology Partners practices on the reporting product. Since June 2026, outside hospitals and health systems can buy it too.
The drafting research covers chest, abdominal and musculoskeletal X rays plus head, cervical spine, chest, abdominal and pelvic CT, and the CT study used test groups for each scan type. Mammography, worklist tools and other planned modules sit outside that research, and the reporting product's spread across affiliated practices shows where it runs without measuring results by subspecialty.
Alternatives to MosaicOS
Other products that do the same kind of work as MosaicOS, and what each one does.
Lithuanian radiology AI developer based in Vilnius, part of Sectra, the enterprise imaging company, since 2026.
Generative AI for radiology reporting and follow up.
Harrison.ai, headquartered in Sydney, sells radiology AI that until 2025 was sold under the Annalise.ai name. Harrison.ai combined with Annalise.ai in February 2025 and renamed the Annalise products in October 2025, and the annalise.ai domain now forwards to harrison.ai.
Romanian imaging AI company based in Timisoara whose Rayscape CXR analyzes chest X-rays, detecting and localizing 148 findings across 17 clinical classes, including tuberculosis, with bone suppressed views, an automatic cardiothoracic ratio and a normal or abnormal flag for worklist…
Lunit is a cancer focused medical AI company founded in 2013, headquartered in Seoul and listed on KOSDAQ. Its radiology line is Lunit INSIGHT: MMG for 2D mammography (FDA cleared 2021), DBT for tomosynthesis (cleared 2023) and CXR for chest X-ray (CXR Triage cleared 2021).
CARPL is a vendor neutral platform that lets radiology departments and imaging groups buy, test, deploy and monitor other companies' imaging AI through one integration and one viewer.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Custom quote for Mosaic Reporting and platform deployment
|
Mosaic Reporting is sold to hospitals, health systems, and radiology organizations through a tailored commercial agreement. The platform supports modular expansion as additional applications become available. No per report, per user, or per site rate is specified. | Mosaic's business associate terms for software customers are not stated. | Implementation is planned around organizational requirements, system complexity, and site count; fees are not specified. | Vendor Published |
Commercial reporting software and investigational image drafting have different availability. Cognita's investigational vision language models are not for sale in the United States. A demo request begins the commercial discussion; there is no stated free production tier, trial allowance, minimum commitment, or public subscription schedule. Equipment replacement and integration scope depend on the deployment.