A second read on
every radiograph.
Oralytix flags regions worth a closer look and lines up historical images automatically, so the dentist starts from a shortlist, not a blank film. Every finding is a suggestion until a dentist confirms it.
Dental imaging is information-rich. The workflow around it isn't.
A single visit can produce a panoramic scan, a set of bitewings, and two years of prior studies to check against. None of that is a shortage of information. It's a shortage of time to look at all of it, tooth by tooth, film by film.
- 01Open the patient record and locate the relevant image
- 02Step through each tooth individually, front to back
- 03Flag anything that looks worth a second look
- 04Pull up prior studies and compare by eye
- 05Write it up, then explain it to the patient in plain terms
How a scan moves from upload to chart note
This is the actual path an image takes. Each stage either checks the image or hands a decision back to the dentist.
Image received
Bitewing, periapical, panoramic, or CBCT, pulled in from the sensor or uploaded directly.
Quality check
Exposure, positioning, and cropping are checked before analysis runs, so a blurry film doesn't produce a false-confident finding.
Tooth detection
Individual teeth are identified and numbered, giving every later finding a fixed address in the mouth.
AI analysis
The model looks for candidate findings (caries, existing restorations, periapical changes) across the full image.
Visual highlighting
Candidate regions are outlined directly on the image, not described in a separate report.
Historical comparison
Prior studies for the same patient are aligned automatically, and regions that changed are marked.
Dentist review
Every finding sits in a queue of three actions: confirm, dismiss, or modify. Nothing is written to the chart unreviewed.
Clinical documentation
Confirmed findings are turned into structured notes in the dentist's own phrasing, ready to edit.
Imaging history updates
The patient's timeline updates automatically, so the next visit starts with full context already in place.
Follow one scan through the system
The same four stages from the workflow above, made interactive. Click a stage to see what it actually looks like, including the two seconds where the model does its job.
Step 1: Upload
A bitewing, periapical, panoramic, or CBCT image is pulled in from the sensor or uploaded directly. Nothing is analyzed yet. This step only checks that the file arrived intact.
Click "02 AI analysis" above to see what the model checks.
The finding lives where the eyes already are
Instead of a text line reading "possible caries, lower right," the region itself is outlined on the radiograph. Click it, and a panel opens with the tooth number, the finding, and the model's confidence. The same interaction shown in the scan above.
- Click-through, not scroll-throughThe dentist inspects flagged regions directly instead of scanning the whole image from scratch.
- Confidence shown plainlyEvery finding carries a model-confidence figure, so low-certainty candidates read differently than high-certainty ones.
- Explainable by designBecause findings are anchored to a visible region, they're easy to check against the image, not just take on faith.
The four places imaging review loses time
Each one addresses a specific point where imaging review loses time, without asking a practice to change how it charts.
Findings you can see
Candidate regions are outlined directly on the radiograph, with tooth number and model confidence attached, never a text line in a separate report.
Learn moreThe dentist has the final say
Every AI finding sits behind three actions: confirm, dismiss, or modify. Nothing reaches the patient chart unreviewed.
Learn moreEvery study, one timeline
Prior radiographs line up automatically by date, turning two years of loose files into one continuous view.
Learn moreRetake before you misread
Exposure, positioning, and cropping are checked before analysis runs, so a blurry film gets flagged instead of confidently misread.
Learn moreBuilt around the parts of the review that eat the most time
AI image analysis
A first pass across the full image returns a short list of regions worth a closer look, not twenty flags, just a handful that matter.
Automatic tooth detection
Every tooth is identified and numbered, so a finding, a note, or a prior treatment always has a fixed home to attach to.
Imaging history, laid out by date
Every prior study lines up on one timeline, so the dentist is comparing against what actually happened last visit.
Compare mode: side-by-side, overlay, or difference
Pick two studies and Oralytix aligns them automatically. A note like "potential change, region of tooth 36" points to exactly where to look. The dentist still makes the call.
Image quality assistant
Checked before analysis runs: exposure, positioning, cropping, and blur. A flagged image gets retaken instead of confidently misread.
Confirm, dismiss, or modify
Every AI finding sits behind a decision. Nothing reaches the chart without a dentist's sign-off.
Notes in your shorthand, expanded into the chart
A quick note becomes a structured entry the dentist edits before saving, so three dentists writing three different shorthands still produce consistent charts.
The same finding, explained twice
A radiograph is easy for a dentist to read and hard for almost everyone else. Patient mode keeps the technical view for the chart, and swaps in plain language for the conversation across the room. The dentist chooses what the patient sees.
Radiolucency, distal surface, tooth 46. Consistent with early-stage carious lesion. Recommend restorative evaluation.
AI assists. The dentist decides.
Oralytix doesn't diagnose, and it doesn't get the last word. It's built to make the dentist's read faster and less easy to miss something in, never to replace the read itself.
- Every AI finding is labeled as a candidate for review, never a diagnosis.
- Nothing is written to the patient chart until a dentist confirms, edits, or dismisses it.
- Model confidence is always visible next to the finding, not hidden behind a plain "detected" label.
The timing isn't incidental
-
1
Caries detection is validated research, not a bet
AI-assisted caries and abnormality detection is one of the more thoroughly studied problems in dental imaging. The science exists to build on, not invent from scratch.
-
2
GPU inference is cheap enough to run per-scan
Serving a model per radiograph, at seconds-not-minutes latency, is now a reasonable cost. It wasn't five years ago.
-
3
Practices are already digitizing imaging
Digital sensors and PACS-style storage are now the norm in most practices, not the exception. The raw material an AI review needs already exists in the workflow.
-
4
Imaging AI is entering clinical use, with a regulatory path
Cleared dental AI systems already return visual findings like bounding boxes and segmentation outlines. The regulatory category exists and is being actively used, not theoretical.
Each phase earns the next one
Each phase only starts once the one before it has been trusted in real practices. Nothing here is a feature list. It's the order things have to happen in.
Prove the read
The first release stays narrow on purpose: caries and abnormality candidates, one of the more thoroughly studied problems in dental imaging AI. Every finding routes through confirm, dismiss, or modify before it touches a chart. This phase is about earning trust one scan at a time, not covering every condition on day one.
Prove the pattern
Once the first-pass read is trusted, historical comparison stops being two images side by side and becomes automatic change detection, flagging what's actually different, not just what's there. Structured summaries start drafting themselves from confirmed findings, and model performance gets benchmarked across practices, not just within one.
Prove the system
With a validated read and a validated pattern, the model earns a harder job: flagging which patients are due a closer look before their next visit, not just reading the scan already in front of it. Support extends to CBCT and deeper integration with the practice's existing management software, so the review lives where the rest of the chart already does.
Built for medical-image volume and real inference load
Analyzing full-resolution radiographs at practice volume, with a latency budget measured in seconds, is a cloud-compute and GPU problem before it's anything else. That's why the stack is built from purpose-fit services across providers, not one vendor's bundle.
Cloud infrastructure
GPU compute
Identity, documents & delivery
AWS, NVIDIA, and Microsoft Azure are trademarks of their respective owners. Oralytix is not affiliated with or endorsed by them.
A workflow problem at the scale of the entire industry
The imaging review bottleneck described earlier isn't specific to one practice. It exists everywhere dental radiographs get read. Oralytix is sold as B2B software: the buyer is the practice owner or DSO administrator, not the patient.
Solo and small practices
The majority of U.S. dental practices are still solo or small-group, practice ownership has declined from 85% to 73% since 2005, but most reads still happen without a second set of eyes. This is where a first-pass review changes the workflow the most.
DSOs and group practices
Dental support organizations now account for a fast-growing share of U.S. practices and lean on standardized workflows across many locations. Consistent, structured findings are what makes an AI review scale across a chain instead of one office.
Target users
- General and pediatric dental practices
- Dental support organizations (DSOs)
- Practice management software partners
Revenue model
- Per-seat monthly subscription, priced by chair count
- Usage-based pricing for high-volume DSO imaging centers
- Integration fees for practice-management-software partners
Growth plan
- Start with a small cohort of solo and small-group practices
- Expand within a single DSO network once the workflow is validated
- Then expand by region, following where imaging is already digitized
Live, early, and specific about it
The MVP is live in production. This isn't a beta or a waitlist page. These are the real numbers behind it, shown as ranges because that's what's actually verifiable at this stage.
10–30 active MVP users
Dentists and practice staff using the live product today: uploading scans, reviewing AI findings, confirming or dismissing candidates.
3–5 design partners
Practices actively shaping the roadmap, their feedback is what decided the order of the phases on this page.
2–3 pilot commitments
Practices that have agreed to run a structured pilot once the next review milestone ships.
5–8 letters of intent
Practices ready to adopt Oralytix once their pilot criteria are met, not verbal interest, signed intent.
Figures are shown as ranges because exact counts shift week to week at this stage, we'd rather publish a verifiable range than a precise-looking number we can't stand behind. Named partners appear only once they've agreed to be named. The Trust & Review page covers our approach to this in more detail.
The people behind the first version
Oralytix is built by a small team that would rather ship one well-reviewed capability than ten unreviewed ones.
Ezeagwuna Chinemelu Hillary
Focused on making sure the product fits how a practice actually charts, not how a slide deck says it should, and that every partnership starts from what dental teams actually need.
Amara Chukwuemeka
Built computer-vision models for medical imaging before turning to dental radiographs specifically. Owns the line the product won't cross: no finding reaches a chart without a dentist's review.
Who we're talking to, honestly stated
Oralytix is pre-pilot. Rather than publish quotes we can't verify, here's what the current conversations actually look like.
A general dentistry practice evaluating fit for a high-volume bitewing and panoramic workflow.
Solo practice, early access conversation
A multi-chair clinic weighing the review-and-approval workflow against their current charting process.
Four-chair clinic, early access conversation
Interested in patient mode specifically, for practices with a high share of pediatric and first-time patients.
Associate dentist, exploratory conversationNo names are published here until a pilot is confirmed and the practice has agreed to be named. This is the same standard the traction section above holds to.
A note on how this is classified
For practices evaluating the product
Software that analyzes radiological images falls within a medical-device category in the United States, and a number of AI-assisted dental imaging tools, including ones that return visual findings like bounding boxes or segmentation outlines, are already in clinical use under that framework.
Oralytix is built around that reality rather than around it: findings are presented as candidates for review, not as diagnoses, and every one requires a dentist's decision before it becomes part of the record.
See the review flow with your own image types
A walkthrough runs through a real bitewing, periapical, and panoramic study, no patient data required to see how it works.