The Detection of Sleep Apnea Hypopnea Syndrome based on Improved BP Neural Network

The prevalence of sleep apnea hypopnea syndrome (SAHS) is increasing year by year and up to 14% in 2016. The disease poses a threat to human sleep safe. To detect the disease effectively and pre-alert, a method for the diagnosis of SAHS using respiratory signals was proposed. According to the characteristics of respiratory signal, nonlinear characteristics fractal dimension and sample entropy were introduced based on time-domain characteristics variance and clinical zero passing numbers as well as frequency-domain characteristics wavelet coefficients and wavelet energy. Then a 6-dimension feature vector was structured and input into the support vector machine (SVM), back propagation neural network (BPNN) and improved back propagation neural network (IBPNN) based on morphology. The method was verified by the 8 sets data including 3840 samples of the Physionet Apnea Database. Experimental results showed that classification accuracy of SVM and BPNN was 71.2% and 84.5% respectively. The IBPNN based on morphology improved the 85% of classification results and increased accuracy by 7.3%. It is very effective to detect the SAHS using the feature vector and IBPNN based on morphology proposed in this paper and provide an efficient and convenient method for the diagnosis of diseases.