MediView XR

The grade describes the mechanism, not the quality of the product, and this is a case where the distinction matters. What XR90 does is spatial registration and rendering: aligning a prior CT volume to a live ultrasound stream and to a tracked instrument, then displaying the result stereoscopically in the correct anatomical position. That is a hard computer vision and computer graphics problem, and solving it well is the achievement, but it is not primarily a machine learning one.

Segmentation of the CT into the 3D anatomical model is the step most likely to use learned models today, and it is upstream of the differentiator rather than being it. No machine learning component, model or method is named anywhere in public materials, while the company's funding communications describe the platform in terms of augmented reality AND AI. A buyer evaluating this should understand they are purchasing a cleared augmented reality navigation device, and should ask specifically which components are learned and how they were validated.

The constraint here is unusually firm because a regulator wrote it. XR90 is cleared for ADJUNCTIVE use, meaning it supplements rather than replaces standard imaging and technique, and the clinician retains the instrument and the decision throughout. The system displays; it does not act, advise or move anything. That is a stronger structural position than most of this index can claim, and it is enforced by the label rather than by vendor policy.

Held at B rather than A because the safety question in navigation is not autonomy but ACCURACY, and no registration error tolerance, drift behaviour or failure mode is published. A navigation system that is confidently wrong by a centimetre is more dangerous than one that declines to display, and nothing states what the system does when registration degrades, when the patient moves, or when ultrasound and CT disagree.

The capability set is described clearly and the method behind it is not. Public materials name the components a user sees, namely CT derived 3D holographic anatomy, live ultrasound projected in anatomical position, and a holographic light ray tracking the instrument path, but no registration method, tracking approach, accuracy specification or model description is given. For a navigation device the decisive technical disclosure is registration accuracy under stated conditions, and it does not appear in any public material located.

No statement on patient information handling, retention, or whether procedural data leaves the institution was located, and no model, hosting arrangement or sub processor list was named. Two questions here are specific to the hardware and this index has not had to ask either before. The first is environmental capture.

A headset of this type builds a spatial map of the room in order to place holograms correctly, and its outward cameras see the clinical team and the patient, so spatial mapping data is a three dimensional reconstruction of a clinical space rather than a technical artefact. Nothing states whether any of it is retained, whether video or spatial data is used for support, training or product improvement, or whether staff are told what the headset captures while they work near it.

That last is a worker question as much as a patient one, and the staff in the room are not the customer and did not choose the device. The second follows from the remote collaboration capability, which lets a clinician at another location observe and guide during a live procedure.

That is a genuine design strength for understaffed and rural sites and is credited as such, and it also means procedural imaging and whatever the headset is sharing travel outside the institution while the procedure is happening. Ask exactly what is transmitted, over what path, whether the session is recorded at either end, what the remote participant's institution retains, and who is responsible for that copy.

Substantially better evidenced than most of this list, and short of the top grade only on publication. The regulatory evidence is real: a 510(k) clearance carries a reviewed submission behind it, and the indication is specific rather than vague. A trial of the system's navigation accuracy was reported in March 2024 and described as demonstrating precision sufficient for clinical use.

First clinical use is documented with a named operator and a named institution, Dr Bradley B. Pua, Chief of Interventional Radiology at NewYork-Presbyterian / Weill Cornell Medical Center, performing a mediastinal lymph node biopsy in November 2023. The origin is Cleveland Clinic intellectual property with an early clinical investigation team there.

What is missing: no peer reviewed publication was located, the accuracy trial's numbers, cohort and methodology are not public, and there is no comparative outcome evidence showing that augmented reality navigation changes procedural success, complication rates or radiation exposure against conventional guidance, which is the claim a health system would need to justify the purchase.

No statement on protected health information handling, retention, or whether procedural data leaves the institution was located. The product ingests a patient's full CT volume and renders it in the procedure room, and augmented reality headsets are capable of capturing the surrounding environment including staff and patient, so there is a real question about what the device records rather than merely displays. Nothing public addresses it in either direction.

A second capability makes that question larger than it first appears. The platform supports real time remote collaboration, with shared visualisation and the ability for a clinician at another location to observe and guide during a live procedure. That is a design strength for understaffed and rural sites and it is credited as such. It also means procedural imaging, and whatever the headset is sharing, travels outside the institution while the procedure is happening. The questions follow directly: what exactly is transmitted, over what path, whether the session is recorded at either end, what the remote participant's institution retains, and who is responsible for that copy.

The environmental capture question deserves its own answer rather than being folded into the first. A headset of this type builds a spatial map of the room to place holograms correctly, and its outward cameras see the clinical team and the patient. Spatial mapping data is a three dimensional reconstruction of a clinical space. Ask whether any of it is retained, whether video or spatial data is used for support, training or product improvement, and whether staff are told what the headset captures while they work near it.

No compliance statement or business associate agreement terms were located. As a cleared device sold into United States hospitals, MediView will be operating under institutional agreements, but nothing publicly verifiable was found, and the practical question is less whether an agreement exists than how many parties the chain has to reach.

Three parties sit in this deployment beyond the vendor and the hospital. The headset platform belongs to Microsoft, which places a third party in the path of data rendered and potentially captured in the procedure room. The company describes partnerships with major imaging manufacturers, so a buyer should establish whether any procedural or imaging data flows to them. And the remote collaboration feature means a clinician at a different organisation can participate in a live procedure, which raises the question of whose agreement covers that participant, their institution and their local recording capability.

That last one is the least likely to be handled by a standard template. A remote collaborator is not the vendor and is not the customer's workforce, so an agreement between hospital and vendor does not by itself govern them. Establish the arrangement before enabling remote sessions across organisational boundaries, and ask specifically whether the vendor's terms contemplate multi institution use at all.

No attestation, trust centre or security documentation was located publicly. Note that a cleared device carries quality system obligations under FDA regulation that a pure software vendor does not, so the underlying engineering controls are likely more formalised than this absence suggests, but nothing verifiable is published.

Two things specific to this product belong on the row.

The first is the host platform. XR90 runs on Microsoft's HoloLens 2 headset, which means the headset's own hardware security features, its device encryption and its platform certifications belong to Microsoft rather than to this vendor. That distinction is easy to elide because the headset is the entire user facing surface of the product, so a buyer looking for assurance will find Microsoft's material first and may credit it here. Establish which layer each control sits in, and what this company is responsible for above the platform.

The second is continuity, which is a security and procurement question at once. This is a cleared device dependent on a single third party headset, and Microsoft has publicly stepped back from that product line. Establish the hardware availability and support horizon directly, ask what the migration path is if the platform is retired, and ask what a change of headset would require of the clearance, since a substantially different hardware platform is not a free substitution for a cleared device. A hospital adopting this is taking a position on another company's hardware roadmap.

The strongest regulatory position on this sourcing list and one of the cleaner ones in the index. XR90 holds FDA 510(k) clearance granted in July 2023 with a specific and legible indication: adjunctive use in minimally invasive ultrasound and CT guided needle based procedures for soft tissue and bone.

The company states it was the first 510(k) clearance for an augmented reality device combining live imaging with 3D extended reality visualisation across both pre operative and intra operative indications, which if accurate places it in the category creating class this index recognises in Viz.ai's first computer aided triage De Novo and Cognivue's De Novo. Two points of discipline.

The clearance covers XR90 as an adjunctive visualisation and navigation aid, so it authorises the display, not any autonomous determination. And a buyer should read the indication against their intended use, since needle based soft tissue and bone procedures is narrower than surgical navigation generally.

No fairness, subgroup or performance variation disclosure was located, and there is a concrete version of that question specific to this device rather than a generic one. Registration accuracy between a prior CT and live ultrasound depends on tissue characteristics, body habitus, respiratory motion and acoustic window quality, all of which vary systematically across patient populations.

A navigation system that registers less accurately in larger patients or in those with poorer acoustic windows would be quietly less safe for those patients, and no per population accuracy data is published. Separately, to record neutrally: Mayo Clinic and Cleveland Clinic are investors in the company, Cleveland Clinic is also the origin of the underlying intellectual property and hosted the early clinical investigation work.

That is common in device commercialisation and is not an accusation, but it means the institutions closest to validating the technology also hold an interest in it, the same caveat this index records for CodaMetrix.

The capability set is described clearly and the method behind it is not. Public materials name the components a user sees, covering derived three dimensional holographic anatomy, live ultrasound projected in anatomical position, and a holographic guide tracking the instrument path, and no registration method, tracking approach, accuracy specification, evaluation or warranty, indemnity or remediation commitment was located.

For a navigation device the decisive disclosure is registration accuracy under stated conditions, and it is absent, which matters more here than a missing accuracy figure would for a detection aid. A detection tool that is wrong produces a flag a clinician can disregard; a navigation overlay that is wrong produces a needle in the wrong place, and the error is invisible at the moment it matters because the display shows the instrument exactly where the system believes it to be.

Registration also degrades with the specific things that happen during a procedure, including patient movement, respiration, tissue deformation as an instrument advances, and drift between the preoperative volume and the present anatomy, so accuracy is meaningless without the conditions attached.

Ask for registration accuracy with the conditions stated, how drift is detected and displayed, what the system does when confidence falls, and whether a clinician is shown an uncertainty margin rather than a single line.

Interoperability here means imaging systems rather than the electronic health record, and on that side the product is necessarily connected: it consumes prior CT volumes and ingests a live ultrasound feed intraoperatively. Neither the imaging standards supported, the ultrasound systems validated, nor the PACS or CT vendors integrated are named publicly, and no EHR or procedural documentation path is described. The GE HealthCare investment raises an obvious question a buyer should ask directly, namely whether integration is being optimised for or restricted to particular imaging estates.

On premise in the procedure room by necessity, since the system must register to live ultrasound with no tolerance for latency. Display is delivered through the Microsoft HoloLens 2 headset, and that dependency is the most important deployment question a buyer should raise.

The clinical viability of a cleared device rests on hardware whose roadmap, availability and support lifecycle belong to a third party and sit outside MediView's control, and a headset that becomes unavailable or unsupported is a supplier continuity problem for the platform rather than a peripheral one. This index treats supplier continuity as a legitimate procurement criterion on the Behold.ai precedent.

Establish the hardware continuity plan and whether XR90 is being ported to alternative headsets, and note that a change of display hardware may itself carry regulatory consequences. No hosting, region or data residency detail is published.

No price, tier or commercial model is published. For a device platform the relevant questions are more numerous than for software and none is answered publicly: whether the system is capital purchase or subscription, whether headsets are included or separately procured, whether per procedure consumables or tracking hardware apply, and what service and calibration coverage costs.

Reimbursement is the other half a buyer must model, since augmented reality navigation adjunctive to a needle procedure has no obvious separate payment pathway, and no reimbursement position is published.

Narrow, well defined and honestly stated, which is worth more than a broad claim. The cleared scope is minimally invasive ultrasound and CT guided needle based procedures for soft tissue and bone, with biopsies and tumour ablations named as the exemplar uses, and the primary buyer is interventional radiology rather than the operating theatre. Named clinical settings are NewYork-Presbyterian / Weill Cornell and Cleveland Clinic.

The company describes broader ambitions across image guided intervention and has separately explored applications such as assessing resectability in ovarian cancer, but those sit outside the current clearance. A buyer in orthopaedic, neurosurgical or laparoscopic navigation should treat this as out of scope today.