Comparison of the ANN based Classification Accuracy for Real Time Sleep Apnea Detection Methods

Cafer Avcı and Ahmet Akbaş


Sleep Apnea, Respiratory Signal


Artificial neural network (ANN) based classification accuracy of the real time sleep apnea detection methods are compared in this study. Comparison has been made between the methods depending on the wavelet analysis of the electrocardiogram (ECG) derived respiratory (EDR) signal and directly measured respiratory signals. EDR signal is computed by 0.2-0.8 Hz band-pass filter implementation on ECG signal. Respiratory signals are obtained by nasal, chest, abdominal based respiratory measurements. Both the ECG and respiratory signals are gathered from ploysomnography recordings of the apnea-ECG database on PhysioNet databank. They are analyzed by using wavelet decomposition of the signal segments having the 1-minute and 3-minutes length. Preliminary tests have shown that, the variances of 10th and 11th detail components can be used as discriminative features for apneas. The features obtained from totally 8 recordings are used for training and testing of a feed-forward ANN classifier. For generalization of the ANN, training and testing process have been repeated by using the randomly obtained 5 different sequences of whole data. The results have shown that the best accuracy can be achieved by analyzing the 3-minutes segments of the nasal based measured respiratory signal. In this case accuracy is greater than 94.4%. However the accuracy of ECG derived signal is better than some of the measured signals, depending on the segment length and measuring type. So that, accuracy is greater than 91.5% for the EDR signal.

Important Links:

Go Back