Towards a fully automated tool for annotation of phasic electromyographic activity

Salient muscle activity identification via the phasic electromyographic metric (PEM) in human polysomnograms/sleep studies (PSGs) represent a potential quantitative metric to aid in differentiation between neurodegenerative disorder populations and age-matched controls. A major impairment to the implementation of PEM analysis for clinical assessment of neurodegenerative disorders includes the time consuming aspects for both visual and automated supervised methods, which require exhaustive expert scoring of PEM and non-PEM events. In order to surmount the aforementioned concerns, we propose a semi-supervised classification methodology encased within an easy-to-use graphical user interface (GUI) utilizing an embedded Minimum Description Length (MDL) criterion to automatically classify PEM and non-PEM events based on expert labeling of a single PEM instance. Results indicate that the application of a semi-supervised approach for PEM identification provides an excellent option to reduce the labeling burden within current human PSG muscle activity identification schemes.