Crowd-sourcing of sleep presented in denver

Last week I (Ben Yetton) presented preliminary results at the SLEEP2016 conference in Denver, CO. Feedback was positive!

I want to take the opportunity to publicly answer some of the questions asked during my talk and poster session.

1. Have you considered other sleep features?

Yes! Right now we are using the web interface to crowd-source detection of sleep spindles. The great thing about the interface is its so versatile, and could be used to crowd-source any sleep feature (or any EEG feature for that mater). We plan to move onto Rapid Eye Movements next.  Remember that this project is all open source so if you want to use it to crowd-source for your data then go ahead!

2. Why are the current automatic algorithms for sleep feature detection so bad?

Generally the issue is not with the quality of the algorithms themselves, it's the data used to develop them that is at fault.  Automatic sleep feature detection algorithm are currently trained a small datasets that contain only a small subset of the possible variance in sleep features. When these algorithms are used on real world data, with examples of sleep features that they are not used to seeing, then they perform poorly. 

3. I'm a sleep tech, will this project put me out of a job?

NO! There is no substitute for a high quality human expert scored sleep data. What this project aims to do is help provide materials to augment the work of sleep techs. For large projects containing thousands of nights of sleep data, expert human feature detection is just not feasible. The amount of sleep data is only increasing, we need these automatic detectors, but they will never replace the high quality work of trained polysomnography technologists.

Do you have other questions? Send us a message on Experiment.com or send an email to MODA.sleepscoring(at)gmail(dot)com.

Here is a link to the poster presented at the Denver conference: https://osf.io/ejdkn/