For all subjects: 4 nights with polysomnography and ear-EEG

For a subset of the subjects in arm the '4 nights with PSG', a second phase follows in which each subject sleeps 12 nights with only ear-EEG.

If a night's recording is unsuccessful, for whatever reason, up to 6 additional nights may be attempted.

soft silicone electrode array placed in each ear (outer ear-canal and concha), connected to a battery powered EEG amplifier.

The test outcome is a set of matched polysomnography and ear-EEG sleep measurements. From this will be generated an algorithm for automatic sleep scoring based on ear-EEG (using leave-one-subject-out cross validation). The primary outcome measure of the test is the correlation between the automatically generated hypnograms and those generated manually from the scalp recordings.

The accuracy is quantified using Cohen's kappa, which is a number between -1 and 1. An average (across all recordings) above 0.4 would be a success for the test.

As the training of the sleep scoring algorithm requires large amounts of data, it is necessary to use a large number of subjects (20) to estimate the viability of automatic sleep scoring from ear-EEG recordings. This also means that kappa values are calculated for all recordings at once when the measurements are done.

This method for creating sleep scoring algorithms and quantifying their success is in line with standard procedure in this field.