A comparative analysis between standard polysomnographic data and in-ear-EEG signals

Polysomnography (PSG) is the gold standard to assess sleep disorders, involving overnight recordings of several bio-signals. The PSG setting is uncomfortable and impractical to use at home. Hence, less invasive, cheaper, and more portable solutions are being proposed. In this thesis, the aim is to understand the limitations and potentials of the in-ear electroencephalography (in-ear-EEG). In particular, the assessment focuses on determining which of the standard PSG derivations is more similar to the in-ear-EEG signal. A per-class feature-based analysis is performed on ten healthy participants (18-60 years). Features in both time and frequency domains are extracted from the standard twenty-one PSG derivations and the in-ear-EEG signal. An unsupervised feature selection procedure is performed to first identify the most relevant features, and then statistically compare their distributions (i.e., PSG vs in-ear-EEG features) for each subject and for each sleep stage. Similarity-scores are then assigned to all the investigated standard PSG derivations. Outcomes are finally aggregated on each class and for each subject to provide a ranked list of the closest PSG channels to the in-ear-EEG. Unexpectedly, for each subject different subsets of PSG derivations are found more similar to the in-ear-EEG recordings. The scalp-EEG channels with the highest affinity to the in-ear-EEG are the frontal ones along with the mastoid-to-mastoid derivation. EOG channels surprisingly show great similarity too. In-ear-EEG seems a valuable solution for home-based sleep monitoring, however further studies with a larger and more heterogeneous dataset are needed.