Convolution neural network for identification of obstructive sleep apnea

Identification of patients with obstructive sleep apnea from normal subjects is essential for most of hospitals. Artificial intelligence techniques are encouraged for simplicity and being less costly and also for more accurate performances compared to traditional identification methods in hospitals A convolutional neural network is used in this work for feature matching process, while the continuous wavelet transform is used for feature extraction. 40 obstructive sleep apnea subjects plus 20 normal subjects RRI data are used in this work. The data is obtained from the MIT databases. The data is divided into 80% for training and 20% for validation. A compromise between the data size and the efficiency of identification is studied. The data is divided into different lengths segments for this purposes. The results are shown in terms of subject identification and also in terms of segment identification. Voting process is included to identify subjects based on segments identification results. The best subject identification result obtained is 93.8% for trial group and 83.3% for validation group. The best segment identification result obtained is 88.45 for trial group and 82.5% for validation group. By using voting among segments a 100% identification of both trial and validation groups can be obtained