Video-based pupillometry using Fourier Mellin image correlation.

We introduce a novel method for evaluating the pupil light reflex (PLR) response using digital video recordings. Expensive, specialized devices are replacing traditional penlight tests in emergency and neurotrauma departments, but they are not widely accessible elsewhere. Recently, smartphone applications have emerged as alternatives, but their accuracy and reliability are limited by their dependence on deep learning and machine vision techniques. We propose a correlation-based approach to PLR measurement using Fourier-Mellin Correlation (FMC), which bypasses pupil detection. FMC, a translation-invariant correlation kernel, measures the scale change of an object. We generated synthetic PLR recordings for various eye colors to validate our method. We conducted an error analysis on the constriction ratio (CR) and constriction velocity (CV) measurements using Monte Carlo simulations with a rendering model that simulates pupil behavior under light stimuli. We considered possible error sources, including added noise, magnification change, frame rate, and compression. Our analysis revealed that frame rate had the most significant impact on CV error (bias error: 0.65%, random error: 23.81%), while added noise had the most significant effect on CR error (bias error: 4.83%, random error: 6.38%). Real data from four human subjects, using smartphone cameras with different eye colors, was also collected to demonstrate the method's applicability. The real data showed the potential of this approach to collect biomarkers for PLR response, including maximum constriction value and timing, time to 75% recovery of the initial constriction, average dilation velocity, and average constriction velocity. Results support the robustness of this workflow in detecting pupil dynamics under varying image conditions, suggesting its potential for low-cost, accessible PLR measurement using any video recording device, including widely available smartphones, thereby eliminating the need for additional hardware and complex pupil detection methods. This advancement holds the potential to enhance neurological assessments and patient care. Future studies should focus on expanding sample sizes and incorporating diverse conditions to further validate the method.