Feature extraction of light sleep state in polysomnogram by use of conditional probability

Human sleep stages wring whole night are usually classified into six stages based on polysommnographic (PSG) record. Sleep state of human light sleep changes gradually and continuously. In this study, automatic judgment of light sleep state in PSG record was developed. Parameters for characterizing the PSG were calculated from the periodogram and the discriminant function was constructed by using conditional probability. Estimated value of wake full level was gradually decreased according to the stage charged from W to S1. In contrast, changing the sleep state from S1 to S2 increased estimated value of sleep level accordingly.