Bottleneck HRV Features for Screening Sleep Apnea Disorder

Sleep apnea; a respiratory sleep ailment, where individual feels sudden ceasing in their breath during sleep. This disorder if goes untreated can lead to several critical health disorders. Rural areas, who are often deprived of speciality systems, disorders like sleep apnea remains untreated. In this work bottleneck features (BNF) derived from a convolutional neural network (CNN) model has been used. BNF are efficient in extracting patterns from the input features. Initially a baseline CNN model was build using the time domain heart rate variability (HRV) features (TD) and frequency domain HRV features (FD) extracted from ECG signal. Subsequently a fully connected network was developed using the BNF extracted from respective CNN models. The TD-BNF model showed comparatively good performance with an overall accuracy of 78.27%, sensitivity of 76.01%, and specificity of 79.73%.