A classification algorithm based on spectral features from nocturnal oximetry and support vector machines to assist in the diagnosis of obstructive sleep apnea

The aim of this study is to develop and evaluate an algorithm to help in the diagnosis of the obstructive sleep apnea syndrome (OSAS). Arterial oxygen saturation (SaO₂) signals from nocturnal pulse oximetry were used to identify OSAS patients. A total of 149 SaO₂ recordings from subjects suspected of OSAS were available. The initial population was divided into a training set (74 subjects) and a test set (75 subjects) to optimize and evaluate our algorithm. Support vector machines (SVM) with Gaussian kernel were used to classify spectral features from SaO₂ signals. Several configurations of SVM were assessed by varying the regularization (C) and the kernel width (sigma) parameters. Finally, the selected SVM classifier (C = 3D 235 and sigma = 3D 0.4) provided an accuracy of 88.00% (84.44% sensitivity and 93.33% specificity) and an AROC of 0.921. Our results suggest that the proposed algorithm could be useful for OSAS screening.