A fused-image-based approach to detect obstructive sleep apnea using a single-lead ECG and a 2D convolutional neural network

Abstract

Obstructive sleep apnea (OSA) is a common chronic sleep disorder that disrupts breathing during sleep and is associated with many other medical conditions, including hypertension, coronary heart disease, and depression. Clinically, the standard for diagnosing OSA involves nocturnal polysomnography (PSG). However, this requires expert human intervention and considerable time, which limits the availability of OSA diagnosis in public health sectors. Therefore, electrocardiogram (ECG)-based methods for OSA detection have been proposed to automate the polysomnography procedure and reduce its discomfort. So far, most of the proposed approaches rely on feature engineering, which calls for advanced expert knowledge and experience. This paper proposes a novel fused-image-based technique that detects OSA using only a single-lead ECG signal. In the proposed approach, a convolutional neural network extracts features automatically from images created with one-minute ECG segments. The proposed network comprises 37 layers, including four residual blocks, a dense layer, a dropout layer, and a soft-max layer. In this study, three time–frequency representations, namely the scalogram, the spectrogram, and the Wigner–Ville distribution, were used to investigate the effectiveness of the fused-image-based approach. We found that blending scalogram and spectrogram images further improved the system’s discriminative characteristics. Seventy ECG recordings from the PhysioNet Apnea-ECG database were used to train and evaluate the proposed model using 10-fold cross validation. The results of this study demonstrated that the proposed classifier can perform OSA detection with an average accuracy, recall, and specificity of 92.4%, 92.3%, and 92.6%, respectively, for the fused spectral images.

A Signal Segmentation‐Free Model for Electrocardiogram‐Based Obstructive Sleep Apnea Severity Classification

“…ECG signals were directly used to train models and then classify the OSA severity in ref., [7–11] ECG‐derived respiration (EDR) signals or RR intervals, defined as the intervals between the R peaks of successive QRS complexes, were used in ref., [12–15] and alternatively spectral features of ECG signals were used in ref. [16–18]…”

“…For this sake, considerable efforts have been made to reduce the number of the required signals in a large volume of publications. [4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23] For example, the OSA severity was classified using ECG [7][8][9][10][11][12][13][14][15][16][17][18] or SpO 2 signals alone, [19][20][21] using a combination of ECG and SpO 2 signals, [22] or even using multiple types of signals, i.e., ECG, SpO 2 , chest, and abdominal respiratory movement signals. [23] Recently, wearable and portable tools at home for OSA classification have been developed as an easy-to-use alternative to PSG.…”

“…ECG signals were directly used to train models and then classify the OSA severity in ref., [7][8][9][10][11] ECG-derived respiration (EDR) signals or RR intervals, defined as the intervals between the R peaks of successive QRS complexes, were used in ref., [12][13][14][15] and alternatively spectral features of ECG signals were used in ref. [16][17][18] Most of the above-referred deep learning models were trained using the MIT PhysioNet Apnea-ECG database. [27,28] The Apnea-ECG database contains a total of 70 recordings, one and the other half of which constitute the released and the withheld dataset, respectively.…”

“…As referenced previously, a single type of signal was used in ref., [7–21] the majority of which took ECG signals as input, and mostly used deep learning or convolutional neural network (CNN)‐based models to classify the OSA severity. ECG signals were directly used to train models and then classify the OSA severity in ref., [7–11] ECG‐derived respiration (EDR) signals or RR intervals, defined as the intervals between the R peaks of successive QRS complexes, were used in ref., [12–15] and alternatively spectral features of ECG signals were used in ref.…”

Sleep Apnea Classification Algorithm Development Using a Machine-Learning Framework and Bag-of-Features Derived from Electrocardiogram Spectrograms

“…In this way, k -fold cross-validation repeats training and testing k times, and the final accuracy is the average of the k accuracy values for each iteration. In this study, the k -fold cross-validation with k = 5 was performed at the spectrogram-level instead of the participant-level, as some studies have done [ 16 , 42 , 43 , 44 ]. As evaluation criteria, the accuracy, sensitivity, and specificity parameters were computed for classification performance metrics.…”