Adaptive Reconstruction of Pupil Dynamics During Blink Data Loss: A Physiologically-Informed Exponential Recovery and Noise Modelling Framework

This paper presents a novel, physiologically inspired framework for reconstructing missing pupil data during blink-induced interruptions. Pupillometry has become integral to studies of cognition, emotion, and neural function, yet the frequent data gaps caused by blinks can compromise the accuracy of inferences drawn from pupil measurements. This approach addresses these challenges by introducing an exponential recovery model that dynamically adjusts its time constant in proportion to blink duration, thus capturing the non-linear dynamics of pupil responses more effectively than traditional interpolation methods. A localised noise estimation procedure further enhances realism by incorporating both signal-dependent and baseline noise derived from statistical properties of the data surrounding each blink. Additionally, a Savitzky-Golay filter is selectively applied to the reconstructed intervals, preserving key physiological features while mitigating high-frequency artefacts. To facilitate the implementation of our framework, we introduce PRPIP, a Python package that implements our methodology. This work demonstrates the effectiveness of this method using empirical pupil data from visual tasks, illustrating that the reconstructed signals maintain physiological fidelity across a range of blink durations. These findings underscore the potential of combining physiological principles with data-driven noise modelling to generate robust, continuous pupil traces. By offering a more accurate reconstruction of pupil size, this framework stands to advance pupillometric research in fields ranging from cognitive psychology to clinical neuroscience.