Research, evidence, and
how to collaborate.
For researchers, investigators, and clinicians evaluating PupiLUX for academic or study use. Three zones: the technical whitepaper, the curated clinical-evidence library, and the Sponsored Site Program for running your own validation.
PupiLUX is a measurement and screening tool currently undergoing clinical validation. Not a medical device. Not a diagnostic device. The whitepaper below describes a preliminary, single-site, single-grader pilot validation. Results are reported transparently to establish a baseline for subsequent multi-site, multi-grader studies. For informational and screening purposes only.
Preliminary Technical Validation Report
A 17-subject pilot validation of PupiLUX smartphone pupillometry — three orthogonal pillars, a blinded expert grader, and the clinically-actionable signals that drive bedside triage.
Dr. Ajay Bakshi · R.Tej Health Analytics · April 2026
What this report shows
The validation is designed around the two questions clinicians actually ask at the bedside.
PupiLUX is a smartphone-based pupillometry system designed for bedside screening and triage in neuro-ICU and emergency-room settings. On a 17-subject pilot cohort captured at CFS-Delhi-01, PupiLUX measures absolute pupil diameter with a mean absolute error of 0.52 mm. Its reactivity metrics — constriction percentage, latency, velocity — are mathematically scale-invariant and were validated against a pharmacologically-induced reactivity asymmetry. Most importantly, for the clinically-actionable question of “is there anisocoria, and in which direction”, PupiLUX agreed with a blinded expert grader on 4 of 4 (100%) clean drug-phase frame pairs, with definite grader confidence at asymmetries as small as 0.11 mm.
The recurrent finding of this report is that PupiLUX's two clinically-actionable signals — reactivity and direction of anisocoria — are robustly detected even in the small-pupil regime where the absolute-diameter detector is most error-prone. These are precisely the signals that drive triage decisions in opioid toxicity, organophosphate poisoning, pontine stroke, and other miotic presentations.
Three orthogonal pillars
Each pillar addresses a different question, with its own dataset and methodology. Pillars 2 and 3 are deliberately tested on the small-pupil dataset where Pillar 1 is hardest.
Absolute measurement accuracy
MAE 0.52 mm
Mean absolute error of 0.52 mm across 36 paired frames from 17 subjects, with 95% limits of agreement [−0.73, +1.39] mm. Well below the ≥1.0 mm AANN Clinical Practice Guideline threshold for clinically significant new-onset anisocoria.
Screening-grade, not research-grade. 12.4% mean overestimation bias. Single-site, single-hardware, single-grader pilot.
Reactivity is scale-invariant
8 of 8 timepoints
Reactivity metrics (CP, LAT, T75) are mathematically scale-invariant to absolute-diameter error. Validated empirically on a 9-recording pilocarpine dose-escalation experiment: the expected L > R asymmetry was preserved across all 8 valid timepoints, with R-eye constriction reaching 1.22% (essentially non-reactive) at peak drug effect.
Validated in the small-pupil regime (R-eye BPD 2.58–3.14 mm) where absolute-diameter error is largest.
Anisocoria direction agreement
4 of 4 (100%)
Blinded expert grader vs PupiLUX on clean drug-phase frame pairs: 4 of 4 (100%) direction agreement, definite grader confidence on every pair, at asymmetries as small as 0.11 mm — an order of magnitude below the AANN ≥1.0 mm threshold. Combined set including overlay-bearing recovery frames: 6 of 7 (86%).
One baseline false positive (CFS-201, warmup artifact) was correctly disconfirmed by the grader and reported rather than excluded.
How PupiLUX captures a bilateral PLR
A 7-second recording on standard iPhone hardware. Zero adapters. Zero cloud inference.
Hardware
Standard iPhone (iOS 17+), rear camera + torch. No adapters, no accessories.
Capture
7-second bilateral recording at 1080p / 30 fps. Baseline (2s) → stimulus (1s) → recovery (5s).
Detection pipeline
Three-layer on-device cascade: Apple Vision face landmarks, YOLOv8n-seg pupil/iris segmentation, tracked-position fallback.
Calibration
Iris-ruler method using 11.7 mm white-to-white corneal diameter (Rüfer et al., Cornea 2005).
Bilateral
Both eyes captured simultaneously and split via face landmarks — to our knowledge, the only smartphone pupillometry system that does this.
Confidence floor
Per-frame YOLO confidence ≥ 0.50 required; sub-floor frames are excluded rather than reported as false readings.
Screening and triage, not research-grade
PupiLUX is appropriate for screening and triage use in neuro-ICU and emergency-room settings, where the clinical question is “has something changed since the last exam” and “is there asymmetry, and in which direction”.
It is not a replacement for dedicated quantitative pupillometers in contexts requiring sub-millimetre absolute precision, and it is not a diagnostic device. All interpretation of PupiLUX output remains the responsibility of the treating clinician.
Four peer-reviewed sources
Every threshold and parameter cited in the whitepaper is traceable to a PubMed-indexed publication.
Interrater reliability of pupillary assessments
Olson DM, Stutzman S, Saju C, Wilson M, Zhao W, Aiyagari V
Neurocritical Care, 2016; 24(2):251–257
Barriers to thrombolysis in acute ischaemic stroke: an epidemiological review from a tertiary hospital in the Eastern Cape, South Africa
Pasio R, Maharaj R, Pasio K
African Journal of Emergency Medicine, 2026; 16(1):100945
White-to-white corneal diameter: normal values in healthy humans obtained with the Orbscan II topography system
Rüfer F, Schröder A, Erb C
Cornea, 2005; 24(3):259–261
The additive miotic effects of dapiprazole and pilocarpine
Geyer O, Loewenstein A, Shalmon B, Neudorfer M, Lazar M
Graefe's Archive for Clinical and Experimental Ophthalmology, 1995; 233(7):448–451
The science behind quantitative pupillometry
Peer-reviewed evidence supporting quantitative pupillary assessment in critical care, emergency medicine, neuroprognostication, and toxicology. Not evidence for PupiLUX itself — evidence for the clinical technique PupiLUX brings to the smartphone.
Recommended by every major guideline body
Six international organisations mandate or recommend pupillary assessment in critical care and emergency settings.
National Institute of Neurological Disorders and Stroke
Pupillary reactivity should be documented in all patients — TBI classification update.
American Heart Association
Serial PLR in all comatose post-cardiac arrest survivors — Level 1 recommendation.
European Resuscitation Council / European Society of Intensive Care Medicine
Bilateral PLR absence at ≥72h as key neuroprognostication indicator.
American College of Emergency Physicians
Non-reactive pupils = independent risk factor for severe injury in mild TBI.
American College of Surgeons
Quantitative pupillometry recommended in TBI best practice.
Brain Trauma Foundation
Pupillary assessment is a core component of neurological monitoring in severe TBI.
Three landmark studies quantified penlight error
The penlight exam is not merely imprecise — it misses the findings that matter most.
Reliability of standard pupillometry practice in neurocritical care
Couret D, Boumaza D, Grisotto C, et al.
Crit Care, 2016
n=406 measurements. 50% of anisocoria missed; 39% error rate for small pupils.
Underestimation of pupil size by critical care and neurosurgical nurses
Kerr RG, Bacon AM, Baker LL, et al.
Am J Crit Care, 2016
Multiple phases. Systematic size underestimation; anisocoria and reactivity errors.
Interrater reliability of pupillary assessments
Olson DM, Stutzman S, Saju C, Wilson M, et al.
Neurocrit Care, 2016
n=2,329 assessments. 67% false negative rate for non-reactivity; Kappa = 0.40.
Why the pupil matters — mortality and outcome data
Pupillary findings are among the strongest independent predictors of neurological outcome in critical care.
Do trauma patients with a Glasgow Coma Scale score of 3 and bilateral fixed and dilated pupils have any chance of survival?
Tien HC, Cunha JRF, Wu SN, et al.
J Trauma, 2006
100% mortality with GCS 3 + bilateral fixed dilated pupils. 58% survival with GCS 3 + reactive pupils.
Quantitative versus standard pupillary light reflex for early prognostication in comatose cardiac arrest patients
Oddo M, Sandroni C, Citerio G, et al.
Intensive Care Med, 2018
NPi ≤2 has 100% specificity for poor neurological outcome post-cardiac arrest.
Mortality in severe traumatic brain injury: a multivariate analysis
Martins ET, Linhares MN, Sousa DS, et al.
J Trauma, 2009
Bilateral mydriasis: OR 11.52 for death in severe TBI.
"Talk and Die" syndrome in moderate-severe TBI
Arnaout O, et al.
Various, 2025
2–7% of moderate-severe TBI patients deteriorate after initial lucid interval.
Quantitative pupillometry resists common confounders
Quantitative pupillometry is unaffected by common intoxicants
Jolkovsky EL, Guthrie C, Gililland K, et al.
J Trauma Acute Care Surg, 2022
n=325. NPi (composite reactivity) unaffected by alcohol, benzodiazepines, opioids.
Alcohol delays ER admission in TBI patients
Andriessen TMJC, Jacobs B, Vos PE, et al.
J Neurotrauma, 2012
Median admission delay: 4h 6m (intoxicated) vs 1h 7m (sober).
The scale of unmet need
India's ER and ICU volumes create a massive opportunity for quantitative pupillometry — with unique clinical scenarios that have no existing automated solutions.
Simplifying the use of prognostic information in traumatic brain injury
Brennan PM, Murray GD, Teasdale GM.
Neurosurgery, 2018
India: 2.2M TBI cases/year; 69% are mild (GCS 13-15).
Door-to-CT times across Indian medical colleges
Gupta D, Bhatia R, et al. (IMPETUS Collaborative)
Front Neurol, 2025
23 centres, n=2,018. Median door-to-CT: 95 minutes (guideline: 25 min).
Organophosphate and aluminium phosphide poisoning in Indian ERs
Chaudhary S, et al.
Toxicol Rep, 2021
30–44% of all Indian ER poisonings. 92K pesticide deaths/year.
Run a study with PupiLUX.
R.Tej Health Analytics selects a small number of qualified investigators to conduct clinical research using the PupiLUX measurement platform — at no cost to the investigator. We provide the instrument and the secure data vault. You provide the clinical question.
What each approved site receives
- 1,000 test credits (≈ 1,000 bilateral tests)
- TestFlight access to the PupiLUX Trial app for the PI and up to 5 named co-investigators
- Cloud data hosting of all recordings, measurements, and reports
- Secure 30-day data vault — data is retained for 30 days after your first export, then permanently deleted
- Per-patient consent capture on-device, using PupiLUX's ICF template as a starting point
- Free-text labeling of each test (study arm, timepoint, visit number)
- Data export on demand as CSV, PDF, and image files
What the program does not provide
- Protocol design or study planning
- Ethics committee submissions or IRB correspondence
- Patient recruitment or screening
- Statistical analysis or data interpretation
- Medical writing or manuscript preparation
- Expert annotation of recordings (your study team does this)
- Regulatory consulting
These services are available from commercial CROs. PupiLUX's role is limited to providing the measurement instrument and the secure data vault.
Who qualifies
Approval is granted at PupiLUX's discretion against three criteria.
Legitimate clinical research context
The principal investigator is affiliated with a recognised hospital, research institute, or academic medical centre, and has working access to an institutional ethics committee.
Contributes to the measurement-science base
The population, scenario, or sample size meaningfully extends what PupiLUX can validate.
Does not require diagnostic claims
The investigator agrees that PupiLUX reports contain measurements and reference ranges only — clinical interpretation rests with the investigator.
Regulatory position
PupiLUX is not a diagnostic device. Reports do not contain diagnoses, clinical impressions, or decision support. They contain quantitative measurements and reference ranges drawn from peer-reviewed literature. The investigator is responsible for all interpretation and any resulting clinical decisions. Every report carries this disclaimer prominently.
This positioning is deliberate and not negotiable. Study protocols that require PupiLUX to produce diagnostic output will not be approved.
Apply to the Sponsored Site Program
Email info@pupilux.ai with “Sponsored Site Program” in the subject line, a short description of your proposed study, and your institutional affiliation. You will receive a structured intake form by return email. PupiLUX will review within two weeks.
R.Tej Health Analytics · Delhi, India · For raw validation data and analysis scripts, email info@pupilux.ai.