NeuroPupil: A generalization-first framework for scalable and biologically informative cross-species pupillometry

Area of Science:

• Neuroscience
• Computational Biology
• Medical Technology

Background:

• Quantitative pupillometry offers insights into brain function but faces limitations in accuracy, scalability, and generalizability.
• Existing methods struggle with robust measurements across diverse subjects, behaviors, and imaging conditions.

Purpose of the Study:

• To develop NeuroPupil, a deep learning framework for high-throughput, cross-species pupillometry.
• To achieve robust generalization of pupil tracking across subjects, behavioral contexts, and imaging conditions.
• To enhance downstream biological inference through improved pupil tracking fidelity.

Main Methods:

• Systematic benchmarking of deep learning training strategies and network architectures.
• Utilizing pooled multi-subject training with an optimized U-Net architecture.
• Validation across diverse mouse and human datasets.

Main Results:

• NeuroPupil demonstrates improved accuracy and computational efficiency compared to existing approaches.
• The framework achieves reliable and transferable pupil tracking performance.
• Enhanced pupil tracking significantly improves prediction of cortical activity and preserves clinical diagnostic information.

Conclusions:

• NeuroPupil provides a reproducible and scalable framework for large-scale pupillometry.
• Precise and scalable pupil measurement is crucial for linking pupil dynamics to brain activity and clinical phenotypes.
• The framework facilitates applications in systems and translational neuroscience.