A Novel Statistical Analysis Method Using Neural Network Classifier for Sleep Apnea Identification

According to clinical evaluation, the sleep apnea not only causes sleeping disturbance but also influences the quality of sleep obviously. It leads to drowse sleepily and results in a high accident rate in daily activity. Since sleep apnea is commonly associated with an imbalance in the autonomic nervous system (ANS) and can be represented by heart rate variability (HRV), the HRV analysis based on developed neural network classifier was used to identify sleep apnea syndrome in this paper. The novel statistical analysis method was proposed to analyze the correlation between the ANS and sleep apnea by the public clinical database. The proposed method was applied on the examination of the parameters with and without sleep apnea was conducted. Back-Propagation Network (BPN) was applied as the classifier to identify sleep apnea by the parameters. The developed analysis software which is highly correlated with clinical verified Kubios analysis software (correlation coefficient >0.96) and proves the high reliability of this study. As results, several HRV parameters reveal significant difference as sleep apnea occurred during different sleeping status. Moreover, the BPN classifier was employed to identify sleep apnea events and the results show that its accuracy, sensitivity, and specificity are 70.7%, 67.7%, and 72.8%, respectively. The proposed analysis method was verified herein in this study and can be further applied on not only clinical evaluation with polysomnography in a hospital but also the home healthcare with wearable device.