Real-Time Rodent Pupillometry on an Embedded Platform for Neuromodulation

Pupillometry has recently emerged as a sensitive biomarker for autonomic and cortical activity, offering a non-invasive readout for neuromodulation research. However, existing commercial and deep learning-based pupillometry systems are primarily designed for humans, rendering them unsuitable for rodent experiments due to differences in ocular morphology, fur-induced optical artifacts, and the computational demands of high-resolution imaging. In this study, we present a low-cost, integrated real-time rodent pupillometry system built on an embedded platform. The proposed rule-based pupillometry algorithm was optimized with adaptive ellipse fitting, RGB masking artifact suppression, and greedy tracking, achieving robust performance under infrared illumination without requiring GPU acceleration. By minimizing errors caused by complex fur patterns and rodent-specific ocular features, our approach achieved an 86.0% detection rate in rat pupillometry, substantially surpassing the 63.1% attained by existing approach. The system was validated in vivo through vagus nerve stimulation (VNS) experiments in Long-Evans rats, where dynamic changes in pupil size reliably reflected stimulation intensity. By enabling an effective evaluation of VNS, these findings highlight the utility of our system as a practical preclinical tool and underscore the broader potential of pupillometry as a non-invasive biomarker for neuromodulation.