Pupil-DLC: an open-source deep learning pipeline for scalable, markerless tracking of pupil dynamics across conscious and unconscious states
Pupil-DLC is an open-source, DeepLabCut-based deep learning pipeline that enables scalable, markerless, and robust tracking of pupil dynamics across diverse species and consciousness states, outperforming existing methods in accuracy while offering flexible, reproducible analysis for both mouse and human pupillometry.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Imagine your pupil (the black center of your eye) as a tiny, living window that constantly opens and closes to show what's happening inside your brain. When you are alert, focused, or even dreaming, this window changes size. Scientists have long wanted to measure these changes to understand brain states, but the tools they've used so far are like old, clunky cameras: they are slow, break easily in new situations, and struggle to keep up with the action.
This paper introduces Pupil-DLC, a new, free software tool that acts like a super-smart, tireless digital detective for watching these eye windows. Here is how it works, using simple comparisons:
The "Training School"
To become an expert, Pupil-DLC went to a massive training school. It studied over 20,000 snapshots taken from more than 130 videos of mice. These mice were in all sorts of "moods": awake and active, sleepy, or under the influence of various drugs (including psychedelics and anesthesia). A team of humans carefully labeled every single pupil in these photos to teach the computer exactly what to look for.
The "Two-Tool Kit"
Instead of using just one method, Pupil-DLC comes with a dual-tool kit:
- The General Model (GM): Think of this as a master key. It is designed to work instantly on almost any mouse video recorded with infrared cameras, making it perfect for analyzing huge amounts of data quickly without needing to be customized for every single mouse.
- The Individual Model (IM): This is like a tailor-made suit. If you need extreme precision for a specific experiment, this tool fine-tunes itself to the unique features of that specific session.
Why It's Better
The paper claims that Pupil-DLC is more accurate and reliable than the current "gold standard" tools. It doesn't just guess; it provides a confidence score, acting like a weather forecast that tells you, "I'm 99% sure this pupil is open," or "I'm only 60% sure, so you might want to double-check this part." This helps researchers decide how much data to keep without losing accuracy.
Crossing Species Borders
Perhaps the most surprising trick is its ability to generalize. Without any extra training, Pupil-DLC can look at human infrared eye videos and track pupils just as well as it does for mice. If researchers give it a little extra training on human videos taken in daylight or with different cameras, it can handle those conditions too.
The Bottom Line
Pupil-DLC is an open-source, flexible platform that turns the difficult task of tracking pupil changes into a smooth, automated process. It allows scientists to measure brain states in both mice and humans across a wide variety of conditions—from deep sleep to the effects of drugs—using a tool that is fast, accurate, and built to last.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.