Sparse coding for apnea-hipopnea detection

LESTUSSI, F.; RUFINER, H. L.; DI PERSIA, L. E.; MILONE, D.H.

Oro Verde

Congreso; XIV Reunión de Trabajo en Procesamiento de la Información y Control; 2011

The sleep apnea-hipopnea syndrome is a common disorder which could be detected by the decrease of the oxygen saturation. In this work we propose the use of a sparse representation of the SaO2 signal to help in the detection of this pathology. For train and test we used the SHHS Polysomnography Database which include 1000 polysomnograms. We show that the reconstruction of the SaO2 signal from an optimal estimated dictionary has errors in the order 10-13. The activation coefficients obtained from this dictionary was used to non-linearly regress the apnea-hipopnea marks by training a multilayered perceptron. Then we estimate the apnea-hipopnea marks from a test dataset with very good results, with a correlation coefficient of 81.74%.