Apnea-hypopnea index estimation from spectral analysis of airflow recordings

This study focuses on the analysis of airflow (AF) recordings to help in sleep apnea-hypopnea syndrome (SAHS) diagnosis. The objective is to estimate the apnea-hypopnea index (AHI) by means of spectral features from AF data. Multiple linear regression (MLR) was used for this purpose. A training group is used to obtain two MLR models: the first one consisting of features obtained from the full PSDs (MLR_full) and the second one consisting of features from a new frequency band of interest (MLR_band). Then a test group is used to validate the final model. The correlation of spectral features and MLR models with AHI was compared by means of Pearson´s coefficient (ρ). MLR_band reached the highest ρ (0.809). Four different AHI decision thresholds were used to evaluate MLR_band ability to distinguish the severity of SAHS. The accuracy achieved was higher as the threshold increased (69.7%, 75.3%, 80.9%, 87.6%) These results suggest that the automated estimation of AHI through spectral features can provide useful knowledge about SAHS severity.