Assessment of spectral bands of interest in airflow signal to assist in sleep apnea-hypopnea syndrome diagnosis.

In this work, we analyze power spectral density (PSD) from single-channel airflow (AF) in the context of sleep apnea-hypopnea syndrome (SAHS) diagnosis. PSDs from SAHS-positive and SAHS-negative subjects were compared through Mann-Whitney test to find bands of interest. Thereby, we characterized three spectral bands (BW1-BW3) by their relative power (P(R1)-P(R3)) and established relationships with apneas and hypopneas. Then, the single and joint diagnostic ability of P(R1)-P(R3) was assessed by means of K-nearest neighbours (KNN), Fisher's linear discriminant (FLD), and logistic regression (LR) classifiers. The KNN and LR models, obtained from P(R1)-P(R3), showed the best diagnostic ability after a leave-one-out cross-validation procedure. 87.7%-84.2% accuracy and 0.799-0.853 area under receiver operating characteristics curve (AROC) were achieved, respectively. Our results suggest that the bands of interest we defined are related to apneas and hypopneas and, therefore, can be useful in SAHS diagnosis.