Abstract
This study is an attempt to automatically determine sleep stages using deep learning with deep neural network (DNN). Using overnight sleeping data on five male college students for 32 days, deep learning and evaluation were conducted. Input data are the extracted values for eight items drawn from the polygraph data; labeled training data are composed of seven items evaluated by one inspector according to Rechtshaffen and Kales(R&K) method.
As a result, the concordance rates between the estimation of sleep stage data by deep learning and the evaluation by the inspector were as follows: SW (57.7%), SREM (75.8%), S1 (7.1%), S2 (79.2%), S3 (62.8%), S4 (80.9%), and MT (1.8%), and overall, it was 72.1% (κ = 0.57, P<0.01). In this study, deep learning with DNN was conducted, extracting and using values of eight items from polysomnography data; this method could be applicable for the automatic determination of sleep stages.
