Monitoring of Obstructive Sleep Apnea in Heart Failure Patients

This research aims to develop a non-intrusive system to monitor obstructive sleep apnea (OSA) in heart failure patients. Heart sounds and ECG are used to develop a support vector machine (SVM) based classifier. The RMS energy in wavelet sub-bands are used as feature vectors. Feature reduction is performed to minimize complexity without loss of performance. Data from 17 patients is parsed into two minute epochs and randomly partitioned into training and test datasets. The training set is used for parameter optimization of the SVM algorithm and a test data set is used to estimate the generalization error of the algorithm. The proposed algorithm has a 85.5% sensitivity and 92.2% specificity for the detection of OSA epochs.