PupilMetrics Neuro Academy

Orientation & setup

What the instrument is

PupilMetrics Neuro records the pupillary light reflex (PLR) and turns it into quantitative, repeatable measurements of brainstem and autonomic function. It also carries the full static analysis of both pupils, and the Drug Effect Monitor, which uses the same recording to measure what a CNS-active medication is doing. Every result is stored on the device, so the next visit can be compared with this one.

It is a research instrument. It has not been submitted for FDA 510(k) clearance or CE marking, and nothing it reports is a diagnosis. Every output must be interpreted by a qualified clinician in the context of the full clinical picture.

Who it is for, and who it is not for

The app's manual puts it in one line: a clinical monitoring instrument, built to measure healing, not to make accusations. Every module on drug effects in this course comes back to that line.

Required hardware

Neuro is built around the Dino-Lite digital iriscope, a USB eye-imaging device whose LED ring is both the illumination and the light stimulus. On Windows, install the DNVideoX driver first, and connect the iriscope before you launch the app.

The wired iriscope's recording screen on Windows has an LED quadrant test: tap each quadrant and check which lamps light. A dead quadrant does not show in the trace, and a stimulus you have not checked is a measurement you cannot trust.

The alternatives are a wireless Dino-Lite with the WF-20 adapter and, on Android, the phone's rear or front camera with the phone supplying the flash. Each one reports the frame rate it actually achieved.

Platforms

Licensing

On the licence screen, click Start Free Trial, or enter the key you were issued and click Activate License. A capture that fails is not counted against the 20 trial analyses.

The scan flow

The camera mode selector lists the PLR sources and the Drug Effect Monitor first, and the static capture sources below them. PLR Sessions on the start screen opens the recorded reflexes directly. The patient form asks for name, age and sex. Main complaints and the practice name are optional, and the practice name is remembered between sessions. Enter the age carefully: it chooses the PLR reference ranges and the age band the static pupil is compared against.

Desktop shortcuts

Capturing a usable image

This module is about the still images the static analysis uses. Recording the reflex is module 4. Almost every doubtful static result goes back to the image. A blurred capture does not produce a missing number. It produces a confident wrong one, which is worse.

Capture order

The app captures the right eye (OD) first, then the left eye (OS). Analysis needs both.

Camera sources

The quality gate

Every frame is checked against five criteria at once. An image that fails does not reach analysis.

RED at least one criterion fails, and the live message names it. AMBER borderline: a small adjustment is enough. GREEN every criterion passes and the photo is taken automatically. A separate “not an eye” check runs after capture in every mode, manual included.

Lighting, framing, stability

• Lighting. Diffuse and even. The iriscope's own LED ring is ideal. Extra lamps add reflections on the cornea.
• Framing. Let the iris fill at least half the frame width, centred. A pupil near the edge of the frame costs decentration accuracy directly.
• Distance. About 2–3 cm from the eye with a Dino-Lite. Adjust until the pupil margin is sharp, because the margin is what the zone findings and the P/I ratio are measured from.
• Stability. Brace your hand or the device, and ask the patient to fix their gaze on a distant point. The gate rejects blink frames by itself.

Reflections near the pupil

When a reflection lies closer to the pupil border than 20% of the pupil radius, the app shows a warning on the result card, in the PDF and on the stored scan. The analysis still runs and nothing is withheld, but a border finding on that eye may have been produced by the light rather than by the iris. The warning marks images at risk. It does not catch every affected image.

Phones and imported photos

Hold the phone upright and let the eye fill the frame. Every capture path, imports included, ends in a 4:3 crop step. For an imported photo it matters more than it looks: the detector searches for the iris edge at set fractions of the frame width, so an arm's-length photo with a small iris in the middle puts the edge outside that search. Cropping to the eye brings it back into range. Enlarging the image does not, because the fraction stays the same.

Reading the static measurements

Two pipelines, on purpose

Every image goes through two independent analyses. The classical computer-vision pass works on the full-resolution pixels and finds the pupil and iris boundaries. The ML model sees a 224×224 crop centred on the iris and predicts the P/I ratio only, through a calibrated linear fit (R² = 0.856). The result card says which of the two produced the P/I you are reading. Where the two disagree, the confidence score falls, and that disagreement is useful information in itself.

P/I ratio (pupil ÷ iris × 100)

The millimetre figure is a conversion

The app measures a ratio. The image has no scale reference, so the app cannot measure millimetres directly. Every static millimetre figure is the ratio multiplied by an assumed iris diameter: 12.0 mm for adults and teenagers, less for young children. Real adult irises run from about 10.2 to 13.0 mm. Treat the millimetre column as a convention that makes bands comparable, and the ratio as the measurement.

Pupils get smaller with age, so the result is also compared against an age band:

Zone findings: flattening and protrusion

The pupil margin is compared, zone by zone, with its own median radius. A flattening means the margin is pulled inward at that zone. A protrusion means it bulges outward. Each eye reports up to four of each.

Each finding names its zone, its clock span and its deviation. The manual reads the margin as the balance, at each clock position, between the sphincter's inward pull (parasympathetic) and the dilator's outward pull (sympathetic). That reading, and the territory it links to each zone, is a research-derived theoretical framework, not a validated diagnostic standard. The deviation is measured. What it means is a hypothesis.

Shape, position and symmetry

Anisocoria is graded on the converted millimetre difference: the P/I difference in percentage points × 12.0 mm ÷ 100. It only compares like with like. If one eye's P/I came from the ML model and the other's from the classical pass, the comparison uses the classical value for both, because a mixed pair would measure the gap between two methods rather than between two pupils.

Hybrid confidence and the grade

One score combines capture quality (20%), classical confidence (35%), ML plausibility (20%) and the agreement between the two P/I values (25%). The grade follows from it: A from 85%, B from 70%, C from 55%, D below that. Three caps stop a good average from hiding a bad part: weak classical detection caps the score at 40%, poor capture quality at 50%, and strong disagreement between the two P/I values at 35%.

The pupillary light reflex

A still image shows the pupil at one moment. The reflex shows how the pathway responds: how soon the pupil starts to constrict, how far and how fast it goes, and how it recovers. This is the core of Neuro. PLR results have their own screen and their own history, and they are not merged into the static analysis, because they are a different measurement.

The pathway, and why it matters after head injury

The constriction pathway runs through the midbrain. The sympathetic pathway that dilates the pupil again runs from the hypothalamus to the upper spinal cord and back to the eye. The literature links the reflex to brain injury in three ways:

• Direct brainstem injury. The pretectal and Edinger–Westphal nuclei sit in the midbrain, exposed to the rotational forces of concussion. Mild diffuse axonal injury is associated with a measurably slower latency, sometimes before clinical signs appear.
• Rising intracranial pressure. When the uncus herniates, it compresses CN III. The early sign described is a sluggish, asymmetric reflex, which pupillometry can pick up before the pupil looks abnormal to the naked eye.
• Lost cortical modulation. The frontal lobe damps the reflex over repeated stimuli. Losing that input is associated with excessive habituation or paradoxical sensitisation (module 5).

These are associations from the literature, not verdicts the app reaches. The app measures the reflex. Relating it to an injury is your clinical judgment.

Sources

The standard recording sequence

Every PLR source uses the same sequence, so two recordings a fortnight apart are comparable. The pupil is found in every frame, blinks are bridged, and the trace is low-pass filtered before any metric is taken. The 12-second length leaves room to measure the post-illumination response 6 seconds after the flash ends (module 5).

Why three seconds of darkness

Before the flash, all light is off for a full three seconds. This is not padding. It is what makes the amplitude measurable:

• Rod photoreceptors begin to adapt to the dark, raising retinal sensitivity.
• The dilator reaches its maximum under sympathetic drive.
• The pupil reaches its true dark-adapted diameter.
• So the amplitude is measured from the same starting point every time.

Wall-clock timestamps and the measured frame rate

A USB camera does not deliver its nominal frame rate. Rather than assume one, the app stamps every frame with the wall-clock time at capture and reads those stamps, so the baseline and the response are assigned correctly even when the rate wobbles. The report states the frame rate it achieved and a timing resolution: the 95th-percentile gap between frames. At 30 fps a frame lasts about 33 ms, and gaps of 27–131 ms have been seen when frames drop. Latency differences smaller than one or two frames are not meaningful. Latency-sensitive work needs 60–120 fps.

Every parameter gets its own verdict

Latency, constriction, peak constriction velocity, T75 recovery and PIPR amplitude are each compared with the reference range for the patient's age group (18–40, 41–60, 61 and over) and marked Normal, Borderline or Abnormal:

• Normal: inside the reference limits (the mean ± 2 SD).
• Borderline: outside them, but within 3 SD of the mean.
• Abnormal: beyond 3 SD.

Velocity and PIPR are flagged only when they are too small, and T75 only when recovery takes too long. When no age was entered, the ranges fall back to the 18–40 adult band, and the result says so rather than defaulting silently. For adults 18–40, the limits are:

Each range carries an evidence tag on the report: Published (taken from a peer-reviewed study, Bitsios et al. 1996), Interpolated (the 41–60 band, between two published bands) or Estimated (all velocity and PIPR ranges). Treat a borderline or abnormal flag on an estimated range as a hypothesis, not a finding.

Every PLR millimetre figure comes from one scale: an assumed horizontal iris diameter of 11.5 mm divided by the median iris diameter in pixels. Baseline and minimum diameter, peak velocity and PIPR amplitude scale with that assumption, and the result says when the default was used. Latency, constriction percentage and T75 are times or ratios, so the assumption does not affect them.

The grade

Each recording gets a grade from A to D, built from the constriction amplitude, the latency, the recovery and the detection confidence. If no reflex was detected, or the signal was too weak, the grade is D. If 30% or more of the frames were rejected for quality, the grade drops one letter. A grade of D points to a capture problem before it points to a finding: dim the room, wait for the dark baseline, centre the eye, and record again after a rest.

The PMi index is experimental

PMi condenses the PLR panel into one score from 0 to 5, shaped to read like a familiar neuro-index scale: 5 is fully normal, below 4 is borderline, and below 3 is abnormal. It is auditable by design: the formula, weights and normative values are published in the app source, and every recording returns the score broken down term by term.

PMi = 5 · exp(−0.35 · z̄)

Each parameter gets a directional z-score against age-adjusted norms, which penalises deviation only in the pathological direction. The z-scores are combined as a weighted mean: constriction amplitude and peak velocity weigh most, then latency and T75, and PIPR amplitude least. A term that is missing or below quality is dropped and the remaining weights are re-normalised. At least three usable terms are needed. With fewer, the card reads “unavailable” rather than showing a misleading number, and it is withheld altogether when capture quality is poor.

The Drug Effect Monitor

The Drug Effect Monitor uses the same reflex recording as a non-invasive pharmacodynamic endpoint: objective evidence of what a CNS-active dose is doing, for the physician responsible for that dose. It is built into Neuro, and it also ships as a standalone Android application. It is not for anyone trying to find out what somebody took.

The questions it exists to answer

• Is this patient's opioid dose producing the expected degree of CNS effect?
• Is anaesthesia wearing off as it should after surgery?
• Is the ICU sedation level right, or is this patient over-medicated?
• Has this athlete's pain medication cleared enough for a valid neurological examination?
• Is this psychiatric medication producing measurable autonomic effects?

What it does not report

It does not identify drugs. A small, sluggish pupil is equally consistent with each of these:

• Morphine 15 mg every 4 hours for post-operative pain
• Methadone 80 mg a day for opioid use disorder
• Buprenorphine 8 mg as medication-assisted treatment
• Diphenhydramine 50 mg taken over the counter for insomnia
• Pilocarpine eye drops for glaucoma
• A patient who slept three hours and is profoundly fatigued

Every positive pattern is shown with the list of medications that produce findings indistinguishable from it. There is no display without the confounders.

The protocol

Do not shorten the rest. If the pupil has not recovered, the amplitude falls from one trial to the next, and that looks like habituation but is not.

Three trials and the habituation index

In a healthy, unmedicated brain the constriction amplitude falls slightly across repeated identical flashes. CNS drugs disturb that modulation, which no single trial can show. The three curves are drawn on one timeline so you can compare their shape, onset, depth and recovery directly.

HI = (Trial 1 − Trial 3) ÷ Trial 1 × 100%

The index needs at least two trials. With fewer it reads N/A.

Pattern classification

The trials are averaged and a fixed set of rules is applied, in this order. Resting P/I is the pupil–iris ratio in the dark baseline, and constriction is the amplitude of the response.

The two velocity patterns need at least six valid frames in the 2.5 seconds after every flash. At a lower frame rate they are not offered, and the result says so. Their velocity thresholds are marked provisional: they come from one device's sober baseline recordings (21 of them) and have not been validated against confirmed drug-affected recordings.

Indeterminate is a real and frequent result, not a failure. A rule set that always returns a class is guessing.

What the literature associates with each class

• CNS depressants. Opioids, benzodiazepines, barbiturates, alcohol and cholinergic agents suppress brainstem autonomic circuits. Opioids produce bilateral, dose-dependent pinpoint miosis through the Edinger–Westphal nucleus. Benzodiazepines mostly slow the dynamics, with a longer latency and lower velocity, without marked miosis at therapeutic doses.
• CNS stimulants. Amphetamines and cocaine enlarge the resting pupil through sympathetic action on the dilator, while the reflex itself stays brisk.
• Anticholinergics. Dilating drops, scopolamine, tricyclic antidepressants and some antihistamines block the sphincter: a large pupil with a suppressed reflex, which is what separates them from stimulants.
• Cannabis. THC slows the constriction without fully suppressing it: peak velocity falls while amplitude changes little. Alcohol and sedatives reduce both.

PIPR: drug effect or structural damage?

After a bright flash ends, melanopsin-containing retinal ganglion cells (ipRGCs) keep firing for 5–10 seconds and hold the pupil down. This post-illumination pupil response (PIPR) is measured on a standard PLR recording at 1 s and 6 s after the flash ends. The manual reads it together with the reflex:

This is an interpretive framework from the literature, and its PIPR ranges are estimated. The app measures PIPR and flags it. The Drug Effect Monitor's own trials do not compute it, so take PIPR from a standard recording, and the reading is yours.

The ethical framework, as implemented

• Mandatory medication flags. Every positive CNS pattern comes with its Medication Interference Flags: the prescribed medications that produce the same findings. The patient's medication history is required to interpret it.
• A disclaimer you cannot dismiss. Every results screen states that the tool is not validated for drug screening, law enforcement or employment purposes.
• A pattern, not an identification. The app presents a pharmacodynamic pattern. The interpretation belongs to the physician.
• Kept apart. Drug Monitor sessions, pattern labels included, are stored in their own history, never in the scan database. A name is optional: the monitor does not need an identity to measure a reflex.

How the eye was lit

Each session records its illumination: the iriscope's LED ring, the phone screen's red baseline field, a screen flash only, or the rear torch. A red field lets the camera see the pupil in the baseline without constricting it much, because red light drives far less constriction than short wavelengths. Sessions recorded under different illumination are not comparable, and the history says so instead of drawing one line through them. Older sessions that did not record their lighting show it as not recorded, not as a guess.

History, serial monitoring & exports

Three record types, kept apart

Every completed analysis is saved to the device automatically. There is no save step to forget. Open Scan History with Ctrl + H on the desktop, or from the start screen. It has three tabs: Static, PLR and Drug Monitor. They are kept apart because they are not the same measurement.

PLR Sessions on the start screen opens the recorded reflexes directly. It shows the detection rate at the top: the share of recordings in which a reflex was detected. A low rate is a fact about the setup, and it belongs in view.

Recovery is a trajectory, not a reading

Recovery from brain injury unfolds over days, weeks and months. One ambiguous reading means little. A direction across the file means a great deal. PLR Serial Trends plots each parameter across a patient's sessions, and every point identifies the session and value behind it. The Drug Effect Monitor has its own Serial Trend Monitoring. What a recovering series is expected to show:

• Pupil diameter normalising: autonomic tone returning.
• Amplitude improving: the parasympathetic pathway recovering.
• Anisocoria resolving: symmetry between the eyes restored.
• Habituation normalising: the cortical–midbrain circuit coming back.

Compare like with like

A series is only as good as your consistency. Record with the same source, the same room lighting and the same eye at every visit. For static scans, use the same device, distance and framing: the Serial Scan Timeline plots P/I, ellipseness and hybrid confidence for each eye, and its thumbnail strip puts every point one click from the image that produced it. A series that moves while its confidence falls is telling you about the capture, not the patient.

Exports

Auto-save PDF is off by default. Turn it on in Settings and a report is written at the end of every analysis. Everything is written locally. Nothing is synced, so moving a patient's records to another machine means moving the files yourself.

Limits, habits & troubleshooting

A routine for every recording

1. Enter the age as well as the name. It chooses the PLR reference ranges and the static age band.
2. Connect the iriscope before launching. On Windows, run the LED quadrant test.
3. Dim the room, and give the patient a distant point to look at.
4. Let the dark baseline run in full. Keep the eye centred and still.
5. Read the grade before the numbers. At D, let the patient rest and record again.
6. For the Drug Effect Monitor, take the medication history before you read the pattern.
7. Export the PDF and keep the JSON with it. The history record is already saved.

What has been measured, and what has not

• Not a medical device. There is no 510(k) clearance and no CE mark. Every output is for research and documentation.
• Not validated against a reference pupillometer. There is no method-comparison study with Bland–Altman limits of agreement, no test–retest repeatability (intraclass correlation), no sensitivity and specificity of the flags against a criterion standard, and no stimulus specification in physical units yet. All four would be needed before clinical, rather than research, use.
• Millimetres are converted, not measured. The ratio is the measurement.
• The stimulus matters. Constriction, velocity and PIPR depend strongly on the intensity, wavelength and duration of the light and on adaptation. If your stimulus differs from the one a published range was measured with, the comparison is indicative only. Record your stimulus specification with any dataset meant for publication.
• Estimated ranges are hypotheses. All velocity and PIPR ranges are estimated. PMi is experimental.
• Patterns are not identifications. They describe physiological states consistent with broad drug classes. They do not identify substances and are not diagnostic.

Troubleshooting

Clinical judgment is required

No output should be acted on without a physician's interpretation in the full clinical context. PLR metrics are one input among many. The instrument would rather be measured than believed: if a result disagrees with what you see, look at the trace and the capture before you believe either one.