Hypnogram from accelerometer data

This article documents creation of a hypnogram (http://en.wikipedia.org/wiki/Hypnogram) using purely sleep activity data recorded by the MonBaby (http://mondevices.com/monbaby) sensor. The subject was yours truly and the data measurements were done around Jan 2014.

Typically hypnograms are done using EEG, this is an attempt to do it using 3D accelerometer signal.

Below is a hypnogram of normal, healthy adult copied from the Wikipedia article.

Methodology was to convert raw output of accelerometer into a proxy signal that represents activity – first graph above on the first figure.

Second graph is aggregation of number of abrupt movements plotted over time, time interval for aggregation is 30 minutes. The interval divided into 5 quantiles that represent 5 stages of sleep. They are represented as horizontal dotted lines on a second graph. Anything above the top line is Wake state, everything below bottom, deep sleep. Not sure if this method can detect REM state, but it will be among lower quantiles.

So far those are inconclusive results and the method to derive hypnogram from accelerometer data remains untested. But preliminary results look very encouraging.