Pupil Dynamics in Nystagmus Using Computer Vision and Signal Processing

We present a quantitative analysis of horizontal nystagmus using video recordings and computer vision based pupil tracking. A 21.2-second video of a subject with clinically evident horizontal nystagmus was processed to extract precise pupil positions. Image preprocessing included grayscale conversion, median filtering, and adaptive histogram equalization, followed by contour-based segmentation to identify pupils. The raw position traces were smoothed and baseline-corrected to produce binocular signals for subsequent time- and frequency-domain analyses. In the time domain, we computed slow phase velocity, quick phase rate, amplitudes,and binocular coordination metrics using cross-correlation. Frequency-domain analysis was performed via Welch’s method to estimate the power spectral density (PSD), from which dominant frequency, spectral bandwidth, and power-law decay characteristics were derived. Results reveal characteristic sawtooth-like slow and quick phases, with PSD analysis indicating a power-law decay (alpha equal to 1.66), suggesting dominance of low-frequency oscillations. Phase-plane analysis showed no stable periodic patterns, consistent with the irregularity of nystagmus waveforms. These findings demonstrate that automated video-based pupil tracking, combined with multi-domain signal analysis, can yield objective biomarkers for characterizing nystagmus dynamics and potentially support clinical evaluation.