PupilMetrics — questions, answered honestly.
Each question is answered from two perspectives.
Is PupilMetrics an FDA-cleared medical device?
No — and that distinction matters at the bedside. PupilMetrics is research-use software, not a cleared diagnostic instrument. It shouldn’t be used to make or change clinical decisions the way you’d use a cleared pupillometer. Treat its output as a research measurement, not a diagnosis.
For study purposes this is the honest, expected status for a novel tool. The software is licensed for research and educational use, and every metric it reports is openly defined and reproducible. You cite it as research instrumentation, not as a validated clinical device — which is exactly what a methods reviewer wants to see.
The Dino-Lite iriscope is “FDA registered.” What does that actually mean?
It’s important not to over-read this. FDA registration for a Class I device is an administrative step — the manufacturer lists the device with the FDA and self-certifies it as low-risk. It is not a review of diagnostic accuracy, and it is not clearance or approval. For an imaging scope, the registration speaks to it being a listed, low-risk imaging device — essentially an eye-safety and device-listing matter — not evidence that it diagnoses anything.
That administrative status is still genuinely useful, just for the right reason. A Class I registered imaging device is a listed medical device, which smooths the path with an IRB reviewing a human-subjects protocol — you can state the capture hardware is a registered Class I device rather than an unlisted consumer camera. It also establishes a manufacturer with an FDA establishment registration, a helpful precedent when considering future regulatory strategy for the software. We never claim the registration says anything about diagnostic validity.
Why does the Dino-Lite registration matter for my IRB submission?
Because IRBs scrutinize the safety of anything that touches a human subject — here, light exposure to the eye. Using a registered Class I imaging device lets you point to a device that’s listed and manufactured under FDA establishment registration, a cleaner story for the safety section of your protocol than a generic webcam. It doesn’t guarantee approval, but it removes a predictable question.
From the clinical-safety side, the relevant point is illumination exposure. A device positioned as a Class I imaging scope sits in the low-risk tier for that reason. Still, your IRB — not the FDA registration — is the body that actually clears your protocol, so present the registration as supporting context, not as pre-approval.
What is the PMi, and why should I trust a single 0–5 score?
The point of the PMi is that you don’t have to trust it blindly — you can audit it. The formula, the weights, and the age-normed reference model are published, so any reviewer can reproduce the score from the underlying parameters. That’s the opposite of a proprietary sealed index, and for a methods section that reproducibility is the whole selling point.
Clinically, the value is that it’s shaped like the bedside indices clinicians already read, so it’s interpretable at a glance — but I’d stress it’s explicitly not equivalent to any cleared device’s index, and it’s marked experimental and unvalidated. Use it as a research summary measure, not as a substitute for a validated score.
Do I need special hardware to use PupilMetrics?
No proprietary hardware and no per-patient disposables — which, practically, means no recurring consumable cost per exam. It runs on a standard UVC or webcam camera.
The tradeoff to disclose in your methods is that capture quality then depends on your camera and calibration rather than fixed optics. If you want controlled, repeatable illumination and working distance, the supported Dino-Lite bundle is the way to standardize that across a study — worth it when you need capture consistency across subjects or sites.
Can I use PupilMetrics for a study I intend to publish?
Yes — that’s the primary intended use, and there’s a citation track for exactly this. You cite the software by name and version in your methods section, and in exchange the license for that study can be discounted or waived. A neutral or negative result is completely fine; we only ask for the citation, never a favorable finding.
From a clinical-credibility standpoint, that honesty is the right posture — a tool that only wanted flattering citations wouldn’t be one I’d trust. Publishing neutral and negative results alongside positive ones is what eventually builds the evidence base a clinician would need.
What does the ML pupil-deformation detection module actually do?
It’s the capability that doesn’t exist elsewhere on the market. Most pupillometry looks only at the dynamic light reflex — how the pupil constricts and recovers over time. The deformation module adds a static morphological dimension: an ML model flags sector-level flattening or protrusion in the pupil boundary. Combining static shape analysis with dynamic PLR metrics is the novel methodological contribution — no competing device on our comparison does both. For a study, that means you can report a shape dimension nobody else captures.
I’d frame its status carefully, though. It’s a research feature, not a validated clinical sign. It tells you where the boundary geometry deviates, not what that deviation means diagnostically — those are separate questions, and the second one isn’t answered yet. It’s a hypothesis-generating tool: a way to notice morphological patterns worth investigating, not a finding you’d act on clinically.
How is my data handled? Where do captures and patient info go? (desktop)
PupilMetrics is offline-first, and for an IRB this is the strong version of the answer. Patient information, eye captures, and analysis results are stored only on your own machine — never transmitted to CNRI or any third party. All analysis, including the ML model, runs locally on the device (classical computer vision plus a bundled on-device model — there is no cloud AI). The app carries no analytics, crash-reporting, or telemetry services. Concretely: patient info and analysis results live in a local database in the app’s private support directory; eye images and exported PDF/text reports are saved to local folders you control; and license credentials are held in OS-native encrypted storage. The database stores patient name, sex, age, complaints, scan date, file paths, and the analysis results — all on your disk, under your control.
The important boundary is what leaves the device and when. The app’s only routine internet communication is license activation/validation, which sends your license key, an anonymized machine identifier, app version, and OS name — and, on activation, the licensee’s name and email — but never any patient data or images. Everything else stays put. Data leaves your device only when you explicitly act: exporting, printing, or sharing a report through your operating system’s own share or print dialog, or, on mobile, choosing to save an image to your photo gallery. The researcher — not the software — decides if anything ever moves. Because captures can be de-identified and nothing is uploaded, you’re not relying on us for data protection; you’re relying on your own device security, which is the posture a research tool should have.
How is my data handled in the Android app? (mobile edition)
The Android app is offline-first in the same way as the desktop edition: patient information, eye captures, and analysis results are stored only on your device and are never transmitted to CNRI or any third party. All analysis — the computer-vision pipeline and the bundled ML model — runs locally on the phone; there is no cloud AI and no cloud sync. On Android, captures are held in the app’s private external storage and exported reports go to your Downloads folder, under your control. The app carries no analytics, crash-reporting, or telemetry services.
The one mobile-specific difference from desktop is billing, and it’s worth stating plainly. Because the Android app is distributed through the Google Play Store, your subscription payment is processed by Google Play, not by us — standard for any Play Store app, and it means Google handles the transaction under its own privacy terms. We never see your payment details, and critically, the billing path carries no patient data or eye images — it knows only your subscription status. So there are two distinct, separate channels: Google Play handles payment, and everything clinical — captures, patient info, analysis — stays local on your device and moves only when you explicitly export, print, or share it. Data leaving the device is always your action, never automatic. For the payment side, see Google Play’s own privacy policy.
What is the drug-monitoring module for, and who should use it?
It’s the module that distinguishes the Neuro edition, built for pharmacodynamic study designs — tracking how the pupil light reflex changes under a drug’s effect over time. It’s aimed at researchers studying pharmacological effects on pupillary response, with dedicated capture protocols and handling for the dark-baseline conditions those studies involve. It’s a research instrument for measuring PLR changes, cited like any other method.
The boundary matters here: it monitors pupillary response as a research measurement — it is not a clinical drug-screening or intoxication-testing device, and shouldn’t be represented as one. The pupil is a sensitive readout of certain drug effects, which makes it scientifically interesting, but inferring a specific substance or clinical state from that readout is well beyond what this tool claims. Research use, measuring a physiological response — not a diagnostic call.
How does the PMi compare to the NPi?
The NPi — the Neurological Pupil index — is the established, proprietary score with the deepest validation literature in the neuro-ICU, and it’s trusted precisely because of that track record and its EHR integration. The PMi does not replace it and is not equivalent to it. If your work depends on the validated NPi, use an NPi device; PupilMetrics reports raw parameters and its own experimental score, not the NPi.
The difference is philosophical, and it’s the whole point. The NPi’s formula is sealed — you get the number, not the math. The PMi is the inverse: the formula, the weights, and the age-normed model are all published, so you can audit and reproduce exactly how the score was derived. The shape is deliberately familiar to clinicians; the transparency is deliberately the opposite of proprietary. For research where reproducibility is the standard, an auditable index is a feature the NPi structurally can’t offer. Neither is “better” in the abstract — the NPi has validation, the PMi has transparency, and which you want depends on whether you need a cleared clinical score or a reproducible research one.