Abstract— Obstructive Sleep Apnea-Hypopnea Syndrome (OSAHS) is a sleep related breathing disorder that has important consequences in the health and development of infants and young children. To enhance the early detection of OSAHS, we propose a methodology based on automated analysis of nocturnal blood oxygen saturation (SpO 2 ) from respiratory polygraphy (RP) at home. A database composed of 50 SpO 2 recordings was analyzed. Three signal processing stages were carried out: (i) feature extraction, where statistical features and nonlinear measures were computed and combined with conventional oximetric indexes, (ii) feature selection using genetic algorithms (GAs), and (iii) feature classification through logistic regression (LR). Leave-one-out cross-validation (loo-cv) was applied to assess diagnostic performance. The proposed method reached 80.8% sensitivity, 79.2% specificity, 80.0% accuracy and 0.93 area under the ROC curve (AROC), which improved the performance of single conventional indexes. Our results suggest that automated analysis of SpO 2 recordings from at-home RP provides essential and complementary information to assist in OSAHS diagnosis in children. I. INTRODUCTION Obstructive Sleep Apnea-Hypopnea Syndrome (OSAHS) is characterized by recurrent episodes of partial or complete collapse of the upper airway during sleep [1]. Untreated OSAHS leads to several negative consequences in the health and development of infants and young children, such as neuropsychological and cognitive deficits, cardiovascular dysfunction, and/or growth impairment [2]. A recent report of the American Academy o Pediatrics suggests a prevalence of OSAHS in the range of 1% to 5% [2]. The gold standard test for assessing children with suspected OSAHS is in-hospital overnight polysomnography (PSG) [3, 2], in which children’s sleep is supervised and monitored by the use of multiple sensors. Therefore, PSG is costly due to the need of expensive equipment and trained This work has been partially supported by the Ministerio de Economía y Competitividad and FEDER under project TEC2011-22987 and by Project Cero 2011 on Ageing from Fundación General CSIC, Obra Social La Caixa and CSIC and a grant by Consejería de Educación de la Junta de Castilla y León under project VA059U13. Gonzalo C. Gutiérrez-Tobal was in receipt of a PIRTU grant from the Consejería de Educación de la Juna de Castilla y León and the European Social Fund. D. Álvarez, G. C. Gutiérrez-Tobal, and R. Hornero, are with the Biomedical Engineering Group, E.T.S.I. Telecomunicación, Universidad de Valladolid, Paseo de Belén 15, 47011, Valladolid, Spain (phone: +34 983 423000, ext. 4716, fax: +34 983 423667; e-mails: dalvgon@ribera.tel.uva.es, gguttob@ribera.tel.uva.es, robhor@tel.uva.es). F. del Campo is with the Hospital Universitario Río Hortega of Valladolid, Spain (e-mail: fsas@telefonica.net). M. L. Alonso and J. Terán are with the University Hospital of Burgos (Spain) (mlalonso@hubu.es, jteran@hubu.es). staff, and its availability is limited, resulting in long waiting lists [3, 4]. Furthermore, the use of multiple sensors makes PSG highly intrusive and limits the effectiveness of this methodology, often leading to poor results when used on young children and infants [5]. There is an increasing research on novel methodologies in the context of sleep apnea diagnosis in children, including history and physical examination, respiratory polygraphy (RP), daytime (nap) PSG, and ambulatory PSG [2, 6, 7]. The American Academy of Pediatrics reported that these methods tend to be helpful if patients test positive but have a poor predictive value if results are negative [2]. Therefore, further research is needed. In this regard, automated signal processing could improve the diagnostic performance of screening tests for OSAHS detection in children. Electrocardiogram (ECG) [8, 9], photoplethysmography (PPG) [10, 11] and blood oxygen saturation (SpO 2 ) [1, 3, 4, 12, 13] are commonly used in this context. In this study, SpO 2 recordings from at-home RP were analyzed. We used SpO 2 due to its reliability, simplicity and suitability for children. Previous studies in the context of OSAHS diagnosis in children by means of SpO 2 assessed conventional oximetric indexes [1, 3, 4, 12, 13], common statistics [4, 12] and conventional spectral features [12]. In the present research, first to fourth statistical moments and three nonlinear methods were applied: mean (M1), variance (M2), skewness (M3), and kurtosis (M4), as well as nonlinear measures of irregularity, variability and complexity by means of sample entropy (SampEn), central tendency measure (CTM) and Lempel-Ziv complexity (LZC), respectively. These methods previously achieved high performance in the context of OSAHS diagnosis in adults [14-16]. We hypothesized that these measures could provide useful and complementary information to conventional oximetric indexes in children. Genetic algorithms (GAs) and logistic regression (LR) are proposed for feature selection and classification. LR is a suitable alternative to the reference linear discriminant analysis (LDA) in binary classification problems [14, 15] but its performance has been weakly assessed as a diagnostic tool for OSAHS in children [1]. Similarly, GAs are optimization methods that could improve performance in the subsequent classification stage [14]. However, few studies applied feature selection [12]. II. SUBJECTS AND SIGNALS A total of 50 children (23 boys and 27 girls) were included in the study. All children were suspected of suffering from OSAHS and derived to the Sleep Breathing Statistical and Nonlinear Analysis of Oximetry from Respiratory Polygraphy to Assist in the Diagnosis of Sleep Apnea in Children Daniel Álvarez, IEEE Member, Gonzalo C. Gutiérrez-Tobal, IEEE Student Member, María L. Alonso, Joaquín Terán, Félix del Campo, and Roberto Hornero, IEEE Senior Member