Computers in Biology and Medicine 39 (2009) 88--96 Contents lists available at ScienceDirect Computers in Biology and Medicine journal homepage: www.elsevier.com/locate/cbm Automated recognition of patients with obstructive sleep apnoea using wavelet-based features of electrocardiogram recordings Ahsan H. Khandoker ∗ , Chandan K. Karmakar, Marimuthu Palaniswami Department of Electrical and Electronic Engineering, The University of Melbourne, Parkville, VIC 3010, Australia ARTICLE INFO ABSTRACT Article history: Received 31 October 2007 Accepted 26 November 2008 Keywords: Obstructive sleep apnoea Heart rate variability ECG-derived respiration Wavelet Support vector machines Patients with obstructive sleep apnoea syndrome (OSAS) are at increased risk of developing hypertension and other cardiovascular diseases. This paper explores the use of support vector machines (SVMs) for automated recognition of patients with OSAS types ( ± ) using features extracted from nocturnal ECG recordings, and compares its performance with other classifiers. Features extracted from wavelet decom- position of heart rate variability (HRV) and ECG-derived respiration (EDR) signals of whole records (30 learning sets from physionet) are presented as inputs to train the SVM classifier to recognize OSAS ± subjects. The optimal SVM parameter set is then determined by using a leave-one-out procedure. Inde- pendent test results have shown that an SVM using a subset of a selected combination of HRV and EDR features correctly recognized 30/30 of physionet test sets. In comparison, classification performance of K-nearest neighbour, probabilistic neural network, and linear discriminant classifiers on test data was lower. These results, therefore, demonstrate considerable potential in applying SVM in ECG-based screen- ing and can aid sleep specialists in the initial assessment of patients with suspected OSAS. © 2008 Elsevier Ltd. All rights reserved. 1. Introduction Obstructive sleep apnoea (OSA) is a temporary closure of the up- per airway during sleep when air is prevented from entering lungs. This is typically accompanied by a reduction in blood oxygen satura- tion and leads to arousal from sleep in order to breathe. It is a com- mon sleep related breathing disorder with a reported prevalence of 4% in adult men and 2% in adult women [1]. As well as excessive daytime sleepiness, the fragmented sleep due to OSA can result in poorer daytime cognitive performance, increased risk of motor ve- hicle and workplace accidents, depression, diminished sexual func- tion, and memory loss [2,3]. Undiagnosed OSA is now regarded as an important risk factor for the development of cardiovascular diseases (e.g. hypertension, stroke, congestive heart failure, left ventricular hypertrophy, acute coronary syndromes) [4]. If patients are identi- fied and treated at an early stage of OSA syndrome (OSAS), the ad- verse health effects can be reduced [5]. Therefore, early recognition of subjects at risk of OSAS is essential. However, due to the scarcity of sleep laboratories and long wait- ing lists, the vast majority of patients remain undiagnosed [4]. There- fore, there is a strong need for the use of portable recording in the assessment of OSAS. Consequently, it would be possible to diagnose ∗ Corresponding author. Tel.: +61 3 83447966; fax: +61 3 83446678. E-mail address: a.khandoker@ee.unimelb.edu.au (A.H. Khandoker). 0010-4825/$ - see front matter © 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.compbiomed.2008.11.003 OSAS simply and inexpensively from holter ECG recordings acquired in the patient's home. Cyclic variations in R–R intervals (beat to beat heart rate) of ECG signals [6] have been reported to be associated with apnoea events; this consists of bradycardia during apnoea followed by tachycardia upon its cessation. This pattern had been successfully used to detect patients with clinical symptoms of sleep apnoea [7–9]. Various stud- ies have confirmed that several new methods could possibly recog- nize sleep apnoea from heart rate variability (HRV) changes [10–17]. As the dynamic pattern of HRV with OSAS is by no means station- ary, HRV analysis with wavelet decomposition was reported to be an efficient tool for screening OSAS [11]. Besides HRV analysis, morphological information can be obtained by measuring the variations in the QRS amplitude of ECG signals. As we breathe, the position of the ECG electrodes relative to the heart changes, thus modulating the amplitude of ECG signals. From this fact, surrogate respiration signal can be extracted, which is referred to as ECG-derived respiration (EDR) [18,19]. Analysis of such a signal was found to be useful in apnoea monitoring because absence or attenuation of respiratory effort is caused by obstruction of upper airway [20–22]. A comparative study [12] on different algorithms for apnoea de- tection based on ECG signals reported that the combination of param- eters of HRV and the EDR signal gave the best classification results. Several machine learning techniques, i.e., linear and quadratic disc- rimant model [22], CART method [11], Bayesian hierarchical model