Overjet

Computer vision on radiographs is the entire product. The platform analyzes bitewing and periapical images in real time to detect caries including early interproximal and incipient lesions, quantify periodontal bone level in millimeters, and identify calculus and periapical radiolucency, overlaying findings directly on the image. The company states its models are trained on millions of clinically annotated radiographs, and the separate payer side claims product runs the same analysis against submitted claims.

Positioned as chairside decision support with the dentist retaining diagnosis: findings are presented as visual overlays on the radiograph the clinician is already reading, providing objective measurable data to support rather than replace the final diagnosis. The design strength is that the output is inherently reviewable, since a highlighted lesion on an image can be immediately accepted or rejected against the clinician's own read.

The company also markets patient communication as a use case, which is where buyers should think carefully: an AI overlay shown to a patient during a treatment conversation carries persuasive weight, and the payer side product creates a parallel incentive question.

The FDA pathway supplies the transparency floor, since each cleared claim required performance data against a predicate, and the company reports multiple clearances across distinct detection and measurement functions. It states its models were trained on millions of expertly annotated radiographs and publishes commentary on accuracy varying across pathology types, imaging modalities, and real world practice conditions, which is an unusually candid framing. Model architecture and per claim performance figures are not published in full.

Nothing identifies any party in the chain: no model or model family, no hosting arrangement and no sub processor list was located in two passes, and no retention period, de identification statement, training policy or deletion term was found. The structural question here is unusual and outranks all of those, because this company sits on both sides of a transaction.

Radiographs captured in a dental practice are that practice's clinical records, and the same company operates a claims review product used by insurers. To be fair about what that does and does not imply: a payer legitimately receives radiographs submitted with a claim, so images reaching the insurer side through the normal claims process is ordinary and expected, and nothing here suggests otherwise. The narrower question is the one that matters.

Whether images or model outputs generated inside a practice deployment, on patients or procedures never submitted to that insurer, are visible to or usable by the payer product would be a different thing entirely, and nothing published rules it out. The corpus question runs alongside, since a company holding radiographs from both sides of the market has an unusually complete collection and nothing states whether either customer's images improve models used for the other. Ask what technically and contractually separates the two product lines, whether practice images can inform payer models, and for a sub processor list per line.

Regulatory clearance plus published validation, with the caveat that most accessible evidence is company authored. The company reports validation studies demonstrating accuracy that meets or exceeds general dentist performance, cites clinical study accuracy rates above 90 percent for detection of caries, bone loss, and periapical lesions, and reports its platform and claims product in use by large dental groups and insurers covering more than 75 million Americans, which is substantial deployment scale. What is thinner is independent peer reviewed evaluation of the deployed product measuring effect on treatment decisions or patient outcomes rather than detection accuracy.

Converted from Not Rated. The prior note identified exactly the right boundary and it is the question a buyer must map before deployment.

No retention period, de identification statement, model training policy or deletion term was located.

The boundary is between the two customer contexts. Radiographs captured in a dental practice are that practice's clinical records. The same company operates a claims review product used by insurers. Whether any image, finding or derived signal crosses from the provider context into the payer context, and under what authority, is the single most important stewardship question on this record, and nothing published addresses it.

To be fair about what is and is not implied: a payer legitimately receives radiographs submitted with a claim, so images reaching the insurer side through the normal claims process is ordinary and expected. The question is narrower and sharper. Whether images or model outputs generated inside a practice deployment, on patients or procedures never submitted to that insurer, are visible to or usable by the payer product. That would be a different thing entirely and nothing rules it out publicly.

The training question runs alongside it. A company holding radiographs from both sides of the market has an unusually complete corpus, and whether either customer's images improve models used for the other is unstated.

Ask what technically and contractually separates the two product lines, and whether practice images can inform payer models.

Converted from Not Rated. No published position was located, and the dual customer model means there are two distinct business associate relationships rather than one.

No agreement, addendum, role statement, subcontractor flow down, breach notification timetable or review cadence was retrieved for either product line.

Both relationships are structurally required. A dental practice is a covered entity, and a vendor analysing its radiographs is its business associate. A dental insurer is separately a covered entity in its own right as a health plan, and a vendor performing claims review on its behalf is its business associate too. The company therefore holds obligations running to two parties who are adverse to one another in the ordinary course.

That is unusual and it is worth a buyer thinking through rather than assuming standard terms cover it. A practice signing an agreement should understand what it permits regarding data reaching the other side of the business, and an insurer should understand the same in reverse. A single template drafted for one relationship will not address it, because the risk is not disclosure to an outside party but disclosure across an internal boundary between two legitimate customers.

Both agreements certainly exist in contracting given the customer base. Neither is public, so neither party can see how the other is bound.

Ask for the agreement covering your side, and specifically for the terms restricting use of your data in the other product line.

Converted from Not Rated. No independent security attestation was located across two differently phrased searches.

No SOC 2 of either type, no HITRUST, no ISO 27001, no trust centre and no penetration testing statement was retrieved.

The clearance point is worth restating because it is the most common substitution error in this category and it applies to this record squarely. The company holds multiple clearances across detection and measurement claims, which establishes that the software performs as labelled for its intended use. It examines the algorithm, the validation and the labelling. It does not examine encryption, access control, code security, monitoring or incident response, and a cleared device from a company with no security programme is entirely possible.

The comparison within this index is direct and unflattering. The nearest peer in dental imaging publishes a SOC 2 Type II with the type named, alongside specific controls and a selectable data jurisdiction, and graded A here. Both companies sell radiographic detection to dental practices, both hold clearances, and only one lets a buyer verify how the images are protected. That is a difference in disclosure rather than necessarily in practice, and it is the difference this axis measures.

The dual customer model raises the stakes, since a breach would expose both practice clinical records and payer claims material.

Ask for the attestation, its type and period, and for the separation controls between the two product lines.

Among the strongest regulatory positions in the index and the deepest in dentistry. The company received clearance in 2021 for bone level measurement in radiographs with periodontal disease, described as the first clearance of its kind, followed by a second clearance for caries detection and outlining, and it reports a total of seven FDA clearances spanning caries, calculus, bone level, and image enhancement.

The company describes the platform explicitly as software as a medical device operating chairside. Buyers should still confirm which specific cleared claims their deployment covers, since clearances here are function by function rather than platform wide.

No formal governance program was located, but the company publishes unusually frank commentary acknowledging that dental AI accuracy varies significantly across pathology types, imaging modalities, and real world practice conditions, which is the substance a governance disclosure would be expected to address.

What is absent is any published analysis of performance variation across patient demographics, sensor and imaging equipment types, or practice settings, all of which are plausible sources of drift for radiographic models.

The regulatory pathway supplies the floor and the company adds a candour that most vendors avoid. Each cleared claim required performance data against a predicate, and multiple clearances span distinct detection and measurement functions, so the obligations for reporting, complaint handling and correction attach and a practice has a route that does not depend on the vendor granting one.

What lifts this above the bare floor is published commentary acknowledging that accuracy varies across pathology types, imaging modalities and real world practice conditions. That is an unusually candid framing and it is a limitation disclosure in substance: it tells a dentist that a figure quoted for one finding on one modality does not transfer to another, which is true of every product in this category and admitted by almost none of them.

A buyer told that performance varies will ask where, which is exactly the question that leads to a useful conversation. Held below the top grade because the answers are not supplied. Per claim performance figures are not published in full, model architecture is undescribed, and no warranty, indemnity or remediation commitment was located, so the candid framing identifies the variation without quantifying it. Ask for sensitivity and specificity per pathology type and per modality, the practice conditions under which performance was measured, and what the vendor commits to when a finding is missed.

Designed to sit inside existing dental workflow, connecting with existing imaging systems and dental practice management software so analysis appears on the radiograph during the exam rather than in a separate application. The company also operates on the payer side against submitted claims, which is a second and distinct integration surface.

Named practice management integrations and standards support were not located, though the presence of the product as an embedded option inside third party dental practice management platforms indicates a working partner integration model.

Converted from Not Rated. No hosting, region, tenancy or residency terms were located.

The workflow implies the architecture without disclosing it. Real time chairside analysis means an image is captured in the practice and a result returns while the patient is still in the chair, which in a small dental practice with no local compute almost certainly means cloud inference with practice side capture. That is an inference from the product's behaviour rather than a statement from the vendor, and it should be treated as such.

Nothing published states the cloud provider, the region, whether processing remains in the country of care, whether tenancy is separated between practices, what the subprocessor chain looks like, or how long images persist after a result is returned.

One separation question is specific to this record and belongs on this axis as much as on privacy. The company runs two product lines serving counterparties in claims disputes. Whether those run on shared infrastructure, whether a common data store underlies them, and what technically prevents material from one reaching the other are deployment architecture questions with contractual consequences. A logical separation asserted in a contract is a different assurance from physical or account level separation a buyer can verify.

The nearest dental peer publishes a selectable data jurisdiction, so this level of disclosure is achievable in the category.

Ask where inference runs, what persists after a read, and how the two product lines are separated technically.

Converted from Not Rated. No pricing is published for either product line, and the structural fact the prior note identified is more consequential than the missing rate.

The company sells radiographic detection to dental practices and claims review to dental insurers. Those two customers are counterparties in a claims dispute. The practice submits a claim asserting a treatment was warranted; the insurer reviews it and may deny. The same company's models inform both sides of that exchange, and in the dental context the underlying evidence is frequently the same radiograph.

This index has recorded an adversarial pattern before, where one vendor builds counter technology to payer automation while others sell the automation. That arrangement at least placed different companies on each side. Here it is one vendor, which is a sharper version and worth stating plainly rather than implying.

It is not improper and there are legitimate readings. A single consistent standard applied to both submission and review could reduce disputes rather than manufacture them, and objectivity applied evenly is a defensible product thesis. The company does disclose both lines openly.

What is not addressed anywhere public is the obvious question: whether the models are the same, whether calibration differs by customer type, and what separates the two businesses internally.

Ask for the pricing basis on both lines, whether the same model serves both, and what governs information flow between them.

Single specialty by design, which is dentistry, and within it focused on two dimensional intraoral radiography, specifically bitewing and periapical images, covering caries, periodontal bone level, calculus, and periapical radiolucency. Buyer types span individual practices, large dental groups and DSOs, and dental insurers, with reported reach across more than 75 million covered Americans. Three dimensional imaging such as cone beam CT and non radiographic diagnosis are outside the described scope.