Feature selection on single-lead ECG for obstructive sleep apnea diagnosis

Abstract

Many articles that appeared in the literature agreed upon the feasibility of diagnosing obstructive sleep apnea (OSA) with a single-lead electrocardiogram. Although high accuracies have been achieved in detection of apneic episodes and classification into apnea/hypopnea, there has not been a consensus on the best method of selecting the feature parameters. This study presents a classification scheme for OSA using common features belonging to the time domain, frequency domain, and nonlinear calculations of heart rate variability analysis, and then proposes a method of feature selection based on correlation matrices (CMs). The results show that the CMs can be utilized in minimizing the feature sets used for any type of diagnosis.

k-NN-based classification of sleep apnea types using ECG

“…A common approach for sleep apnea detection from ECG is utilizing HRV. The related work based on features obtained from HRV [3,[8][9][10]17,18] seems promising. Nevertheless, employing features from HRV and EDR together seems to be a good practice and provides better classification performance than using HRV features alone [11,14].…”

“…The Physionet apnea-ECG database was most often used by related works [5,[10][11][12][13][14][16][17][18] in which QRS complexes were provided; therefore, no R-wave detection procedure was required in those investigations. In this work, TEO was applied for R-wave detection similar to [28].…”

“…The majority of the related work consists of recordings from a single database [3,[10][11][12][13]15,17,18] and mostly from the Physionet apnea-ECG database [34]. A common approach for sleep apnea detection from ECG is utilizing HRV.…”

“…In the present study, multiple databases were utilized for apnea detection from ECG signals, similar to [5,14,16]; however, the majority of the related work consists of recordings from a single database [3,[10][11][12][13]15,17,18]. The related work is mostly based on features obtained from HRV only [3,[8][9][10]17,18], but in the present study, features from HRV and EDR were employed together, since they provide better classification performance than using HRV features alone [11,14].…”

“…The related work is mostly based on features obtained from HRV only [3,[8][9][10]17,18], but in the present study, features from HRV and EDR were employed together, since they provide better classification performance than using HRV features alone [11,14]. In the related work, HRV and EDR parameters were mostly analyzed in the frequency domain [5,[13][14][15][16].…”