Table 1 Research on detecting obstructive sleep apnea with single lead ECG signals.
From: Integrating physiological signals for enhanced sleep apnea diagnosis with SleepNet
Authors (year) | Techniques used | Preprocessing techniques | Dataset | Accuracy (%) | Advantages | Disadvantages |
|---|---|---|---|---|---|---|
Gutta et al.1 | Vector-valued Gaussian processes | Feature extraction (time-domain, frequency-domain) | PhysioNet Apnea-ECG | 82.33 | Effective for continuous data, robust performance | Limited to specific dataset, not scalable for larger data |
González et al.13 | Binary Classification (FBanks, CC, DFA) | Normalization, feature selection | PhysioNet OSA | 84.76 | Simple, interpretable results | May miss complex patterns due to simplified features |
Li et al.14 | Deep Neural Network and Hidden Markov Model | Time-domain feature extraction, FFT | Custom ECG dataset | 84.76 | High accuracy, deep learning for complex patterns | Requires large training datasets, computationally expensive |
Surrel et al.3 | Support Vector Machine (SVM) | PCA, feature scaling | PhysioNet Apnea-ECG | 88.20 | Efficient for small datasets, high accuracy | SVM sensitive to feature selection |
Papini et al.10 | 10-fold cross-validation | Data normalization | Various clinical datasets | 88.30 | Generalizable across datasets | Requires extensive cross-validation, time-consuming |
Wang et al.15 | Deep Residual Network | Raw ECG signal input, denoising | MIT-BIH ECG dataset | 83.03 | Robust to noise, deep learning approach | May require large data for training |
Singh et al.16 | Pre-trained AlexNet | Data augmentation, feature extraction | PhysioNet Apnea-ECG | 86.22 | Transfer learning improves accuracy, fast computation | Dependent on pre-trained models, may not generalize well |
Feng et al.12 | FSSAE + TDCS classification | Feature scaling, dimensionality reduction | Sleep-ECG dataset | 85.10 | Novel technique for improved classification | Computationally expensive, complex architecture |
Shen et al.2 | Multi-Scale Dilation Attention CNN | Feature extraction, noise reduction | Sleep-ECG dataset | 89.40 | High accuracy, robust to noisy data | Complex model, requires large dataset |
