Gleamer

The Gleamer Copilot suite is a set of detection models and nothing else. BoneView detects fractures, effusions, dislocations, and bone lesions on radiographs, and the portfolio extends to ChestView for chest X-ray and to bone age and musculoskeletal measurement. There is no hardware or services layer; where the software reaches a customer through an imaging platform partner, it is the algorithm being embedded, not a broader product.

Detection with the reader retaining the decision. BoneView presents suspected fractures inside bounding boxes with a three tier confidence label, positive with a solid box above 90 percent confidence, doubt with a dotted box between 50 and 90, and negative otherwise, plus optional worklist prioritization. The graded confidence output is more informative than a binary flag and helps a clinician calibrate trust, but the design keeps the human making the call. Assistive, clearly bounded.

The FDA 510(k) summary for BoneView describes a fully crossed multiple reader multiple case retrospective study with ground truth set by a panel of three US board certified radiologists across all indicated anatomical areas, tested across scanners from multiple manufacturers. That is a specific and inspectable validation design.

Training set size and composition are not published in comparable detail, and there is no third party head to head benchmarking against competing fracture AI, so transparency is strong on the clearance study and lighter on the model internals.

The chain differs by customer here, which is the substantive finding rather than the usual enumeration gap. The software is sold directly and through multiple third party imaging platforms, so the path an image takes depends on which channel a site bought through.

Under a platform deployment there are at least two organisations handling the study, each with its own retention schedule, security posture and sub processors, and a buyer's diligence has to reach both rather than stopping at the company whose name is on the algorithm. Nothing published maps those paths or names the platform partners for a given market. One mitigating fact belongs here for fairness and is easy to overlook.

The company operates under European data protection law at home, which imposes real obligations on purpose limitation, retention and lawful basis regardless of what appears on a website, so the absence of published customer facing terms is not the absence of governing rules.

What is unaddressed is the training question, and the company's own expansion makes it live: models now cover fractures, effusions, dislocations and lesions across adult and paediatric populations, plus chest, bone age and musculoskeletal measurement, and that expansion required data. Ask what is retained after a read, whether customer images train models, whether a site can decline, and which parties are in the path for your channel.

One of the deeper evidence bases among musculoskeletal imaging vendors. The company reports roughly 30 peer reviewed publications, and specific results are documented: a US study through Boston University School of Medicine showing improved sensitivity and specificity with reduced reading time across the appendicular skeleton, rib cage, and thoracolumbar spine, and a 300 patient pediatric study reporting about 91 percent sensitivity that supported the pediatric clearance.

Reported scale is large at 35 million exams across more than 2,000 institutions in 45 countries. The MRMC design underlying the clearance measures reader performance with and without the tool, which is the right question for a detection aid.

Converted from Not Rated. No stewardship terms were located, and the multi party distribution model makes this a per deployment question rather than a single answer.

Nothing was retrieved describing retention of radiographs or derived outputs, de identification, whether customer images contribute to model development, or what is deleted at contract end.

The training question is the live one. The company's models detect fractures, effusions, dislocations and lesions across adult and paediatric populations, and it has expanded into chest, bone age and musculoskeletal measurement. That expansion required data. Whether any of it came from customer images, and whether a site can decline, is unaddressed.

The distribution model compounds it. The software is sold directly and through multiple third party imaging platforms, so the path an image takes differs by customer. Under a platform deployment there are at least two organisations handling the study, with separate retention and security postures, and a buyer's diligence has to reach both.

One mitigating fact is worth noting for fairness. The company operates under the European data protection regulation at home, which imposes real obligations on purpose limitation, retention and lawful basis. Those obligations are not published as customer facing terms, but they are not absent either.

Ask what is retained after a read, whether images train models, and which parties are in the path for your deployment.

Converted from Not Rated. No business associate terms were located, and the distribution model makes the chain the real question.

No agreement, addendum, role statement, subcontractor flow down or breach notification timetable was retrieved.

The company does maintain a published legal and regulatory page carrying its device information, and states data protection under the European regulation, which is its home obligation. That is a real framework and it is not the one a United States radiology buyer needs. General data protection compliance and health privacy rule compliance are different regimes with different instruments, and holding the first says nothing about the second.

The distribution model is what needs settling. The software reaches customers directly and also through several third party imaging platforms, including at least one other vendor held in this index. Where deployment runs through a platform partner, the contracting chain may place the platform between the hospital and this vendor, which changes who holds the business associate agreement, who is liable for a breach and who notifies whom. A hospital buying through an aggregator may have no direct instrument with the company whose model is reading its images.

Ask which entity signs, whether the agreement is direct or flows through a platform partner, and where images are processed.

Converted from Not Rated, and the pattern here is shared across every non United States imaging vendor graded in this pass.

No SOC 2 of either type, no ISO 27001, no HITRUST and no trust centre was located across two differently phrased searches.

What the company publishes instead is a complete device regulatory dossier, and it is genuinely thorough: Class IIa conformity assessed by a named notified body under the European medical device regulation, United States clearances with their submission numbers stated, and a Health Canada licence number. Naming the clearance numbers so a reader can pull the summaries is better practice than most.

The distinction this index draws holds. The quality system standard referenced in device submissions governs how software is designed and manufactured under control. Notified body conformity assessment governs device safety and performance. Neither examines how patient images are protected once the software is running, and neither substitutes for an information security attestation.

The structural explanation is worth stating rather than treating as negligence. A European manufacturer's regulatory effort goes into the pathway that permits market access, while information security attestation is a United States commercial expectation that bites only in procurement, and is typically handled in contract rather than published.

Ask what independent security assessment exists, and whether one is planned for the US market.

BoneView is FDA 510(k) cleared as a computer assisted detection and diagnosis device, cleared K222176, and the company reports it is the only AI fracture detection software cleared for both adult and pediatric use, with a 2026 pediatric clearance for patients over two years of age. ChestView holds a separate FDA clearance. CE marked in Europe since 2020. Clear, current, multi product US authorization with a documented clearance basis.

The clearance study tests performance across multiple anatomical regions and multiple scanner manufacturers, which addresses one form of generalization, and the pediatric study extends validation to a distinct population. What is absent is a formal governance framework, ongoing monitoring commitment, or subgroup performance disclosure across demographics. Some evidence of generalization testing, no published governance program.

The validation design is specific and inspectable, which is what this axis rewards when a commercial commitment is absent. The clearance summary describes a fully crossed multiple reader multiple case retrospective study, with ground truth set by a panel of three board certified radiologists across all indicated anatomical areas, tested across scanners from multiple manufacturers. Each element does work.

A fully crossed design means every reader saw every case with and without the software, which isolates the software's effect rather than confounding it with reader differences. Panel ground truth across all indicated areas means the standard was consistent and no anatomical region was excluded from scrutiny.

And multi manufacturer scanner testing addresses the failure mode that most often breaks imaging models in deployment, where a system validated on one vendor's equipment meets another's and degrades quietly. A buyer can read the design and judge whether it resembles their setting.

Held below the top grade because training set size and composition are not published in comparable detail, no independent head to head benchmarking against competing fracture detection was located, and no warranty, indemnity or remediation commitment attaches. This index credited a competitor in the same cluster for entering multi vendor comparison twice. Ask for the training composition, and whether the company will enter a head to head evaluation.

Distribution breadth substitutes for integration engineering. BoneView is available directly and through a wide set of imaging platform partners including Fujifilm, and other AI orchestration vendors indexed here, meaning a customer can often acquire it inside an imaging workflow already in place rather than as a standalone integration. This is a radiology and PACS environment fit; no EHR integration is claimed or applicable.

Multiple delivery paths exist, direct and through several imaging platform partners, which gives buyers routes to deployment but also means data residency and hosting terms vary by which path is chosen. Specific tenancy and residency terms are not published, and the partner distribution model makes them contract dependent.

No product pricing is published. The company discloses that the US contributes a substantial share of revenue and cites broad deployment scale, but nothing on per exam, per site, or licensing cost, and pricing likely differs between direct sale and partner embedded distribution, neither of which is disclosed.

Centered on musculoskeletal and trauma radiography, where BoneView covers the appendicular skeleton, rib cage, and thoracolumbar spine and is cleared for both adults and children, with the portfolio widening into chest X-ray and bone age. Read by radiologists, orthopedic surgeons, emergency physicians, and others across emergency and outpatient settings. Focused on plain film rather than spanning CT, MRI, and general radiology, which distinguishes it from broad multimodality platforms.