Automated analysis of overnight oximetry recordings by means of support vector machines to assist in the diagnosis of paediatric sleep apnoea

Abstract

Background. Paediatric obstructive sleep apnoea-hypopnoea syndrome (OSAHS) has emerged as a frequent and concerning medical condition in the past 2-3 decades. In-laboratory overnight polysomnography (PSG) is the gold standard diagnostic technique but is complex and relatively inaccessible. Objectives. Blood oxygen saturation (SpO 2 ) from nocturnal oximetry could provide essential information in order to simplify the diagnostic process. The goal of this study was to design and assess an automated classifier aimed at detecting OSAHS. Methods. The population under study was composed of 176 children referred to the Sleep Unit due to suspected OSAHS. All children underwent complete in-laboratory PSG as gold standard. An apnoea-hypopnoea index (AHI) greater than or equal to 5 e/h were considered OSAHS positive. The population was randomly divided into training set (60%) and test set (40%). SpO 2 recordings from PSG were processed offline. Three nonlinear measures were derived from nocturnal SpO 2 recordings and used to design a support vector machine (SVM) classifier. Conventional oxygen desaturation index of 3% (ODI3) was used for comparison purposes. Results. The SVM classifier reached 85.7% sensitivity, 81.0% specificity, 4.50 LR+, 0.18 LR-, and 82.9% accuracy in the test set. On the contrary, ODI3 achieved 89.3% sensitivity, 69.1% specificity, 2.89 LR+, 0.16 LR-, and 77.1% accuracy in the same test set. Conclusions. The proposed SVM classifier outperforms the conventional desaturation index ODI3. Therefore, SVMs and nonlinear measures could provide useful tools to assist in the diagnosis of paediatric OSAHS.