Rodent Sleep Assessment with a Trainable Video-based Approach
Van Anh Le, Mitchell Kesler, Jong M. Rho, Ning Cheng, Kartikeya Murari
Abstract
Assessment of sleep can reveal healthy physiology and behaviour, which are essential to study diseases and treatment. The primary approaches to quantify sleep in animal models are using invasive methods that require implantation of electroencephalogram (EEG) and electromyogram (EMG) electrodes. Those methods are resource-intensive and less than ideal for high-throughput screening. Several studies proposed using video processing to monitor sleep. Those approaches require high quality videos and an optimal threshold value which can be sensitive to different experiment settings. In this paper, we present a trainable video-based approach that can alleviate those limitations. We have come up with a set of effective features at frame-level which are then put into a recurrent neural network to capture long-term temporal features. The result obtained is highly correlated with EEG/EMG-defined sleep.
BibTeX
@inproceedings{icassp2019_rodentsleepasses,
title = {Rodent Sleep Assessment with a Trainable Video-based Approach},
author = {Van Anh Le and Mitchell Kesler and Jong M. Rho and Ning Cheng and Kartikeya Murari},
booktitle = {ICASSP 2019},
year = {2019}
}