Markerless three-dimensional analysis of mouse eye movements using stereo vision

Area of Science:

• * Neuroscience
• * Biomechanics
• * Vision Science

Background:

• * Invasive methods like implanted coils or ocular markers are traditionally used for quantifying three-dimensional mouse eye rotations, including torsion.
• * Existing techniques can be cumbersome and may introduce artifacts.
• * A need exists for less invasive and more practical measurement solutions.

Purpose of the Study:

• * To develop and validate a markerless stereo vision-based method for measuring three-dimensional mouse eye rotations (yaw, pitch, roll).
• * To assess the performance and reproducibility of this novel technique.
• * To provide a practical, non-invasive tool for eye movement quantification in vestibular studies.

Main Methods:

• * Employed a stereo vision system with two geometrically calibrated cameras.
• * Utilized a convolutional neural network for pupil edge detection and ellipse fitting.
• * Reconstructed 3D eye rotations from matched edge points, deriving roll from torsional shifts in pupil-edge irregularities.
• * Validated the method using a 3D eyeball model and in vivo mouse experiments during vestibulo-ocular and otolith-ocular reflex testing.

Main Results:

• * The markerless stereo vision method demonstrated high accuracy in model experiments, with mean absolute errors of 0.05-0.12°/frame.
• * In vivo testing showed reliable measurements of angular vestibulo-ocular reflex gains across varying stimulus frequencies.
• * Static otolith-ocular reflex gains were measured with low variability and consistent rotation axes.

Conclusions:

• * Markerless stereo vision offers a precise and practical approach for quantifying three-dimensional mouse eye movements.
• * This non-invasive technique reduces ocular invasiveness compared to traditional methods.
• * The validated method is suitable for detailed eye movement analysis in mouse vestibular research.