RT info:eu-repo/semantics/conferenceObject
T1 A Bayesian Neural Network Approach to Compare the Spectral Information from Nasal Pressure and Thermistor Airflow in the Automatic Sleep Apnea Severity Estimation
A1 Gutiérrez Tobal, Gonzalo César
A1 Frutos Arribas, Julio Fernando de
A1 Álvarez González, Daniel
A1 Vaquerizo Villar, Fernando
A1 Barroso García, Verónica
A1 Crespo Senado, Andrea
A1 Campo Matias, Félix del
A1 Hornero Sánchez, Roberto
AB In the sleep apnea-hypopnea syndrome (SAHS)context, airflow signal plays a key role for the simplification ofthe diagnostic process. It is measured during the standarddiagnostic test by the acquisition of two simultaneous sensors: anasal prong pressure (NPP) and a thermistor (TH). Thecurrent study focuses on the comparison of their spectralcontent to help in the automatic SAHS-severity estimation. Thespectral analysis of 315 NPP and corresponding TH recordingsis firstly proposed to characterize the conventional band ofinterest for SAHS (0.025-0.050 Hz.). A magnitude squaredcoherence analysis is also conducted to quantify possibledifferences in the frequency components of airflow from bothsensors. Then, a feature selection stage is implemented to assessthe relevance and redundancy of the information extractedfrom the spectrum of NPP and TH airflow. Finally, a multiclassBayesian multi-layer perceptron (BY-MLP) was used toperform an automatic estimation of SAHS severity (no-SAHS,mild, moderate, and severe), by the use of the selected spectralfeatures from: airflow NPP alone, airflow TH alone, and bothsensors jointly. The highest diagnostic performance wasreached by BY-MLP only trained with NPP spectral features,reaching Cohen’s  = 0.498 in the overall four-classclassification task. It also achieved 91.3%, 84.9%, and 83.3% ofaccuracy in the binary evaluation of the 3 apnea-hypopneaindex cut-offs (5, 15, and 30 events/hour) that define the fourSAHS degrees. Our results suggest that TH sensor might be notnecessary for SAHS severity estimation if an automaticcomprehensive characterization approach is adopted tosimplify the diagnostic process
YR 2017
FD 2017
LK http://uvadoc.uva.es/handle/10324/31355
UL http://uvadoc.uva.es/handle/10324/31355
LA eng
NO Producción Científica
DS UVaDOC
RD 22-dic-2024