TABLE 6.
Summary of previously reported early SCD detection using ECG/HRV signals.
| Author (year) | Data (ECG or HRV) | Total no. of features | Method (features) | Classification | Results |
| Early SCD detection using 1-min interval ECG/HRV signals | |||||
| Ebrahimzadeh and Pooyan (2011) | 35 normal and 35 SCD (HRV) Source: Normal Sinus Rhythm (NSR) database and Sudden Cardiac Death Holter (SCD) database | 20 | Linear and non-linear methods (time-domain features (5); frequency-domain features (4); time-frequency domain features (11)) | KNN, Multilayer perceptron (MLP) | Acc = 91.23% (2nd 1 min before) |
| Ebrahimzadeh et al. (2014) | 35 normal and 35 SCD (HRV) Source: NSR database and SCD database | 24 | Linear and non-linear methods (time-domain features (5); frequency-domain features (4); time-frequency domain features (11); non-linear features (4)) | KNN, Multilayer Perceptron Neural Network | Sen = 83.75% Acc = 83.93% (4th 1 min before) |
| Acharya et al., 2015a | 36 normal and 40 SCD (ECG) Source: NSR database and SCD database | 18 | DWT, non-linear methods (non-linear features (6)) | SVM, DT, KNN | Sen = 92.50% Spe = 91.67% Acc = 92.11% (4th 1 min before) |
| Acharya et al., 2015b | 36 normal and 40 SCD (HRV) Source: NSR database and SCD database | 10 | Recurrence Quantification Analysis, non-linear methods (RQA parameters (10)) | SVM, PNN, KNN, DT | Sen = 85% Spe = 88.8% Acc = 86.8% (4th 1 min before) |
| Mirhoseini et al. (2016) | 18 normal and 19 SCD (HRV) Source: NSR database and SCD database | 22 | Linear and non-linear methods (time-domain features (5); frequency-domain features (4); time-frequency domain features (10); non-linear features (3)) | SVM | Spe = 89.5% Acc = 83.24% (1st 1 min before) |
| Fujita et al. (2016) | 18 normal and 20 SCD (HRV) Source: NSR database and SCD database | 9 | Wavelet transform, non-linear methods (non-linear features (9)) | DT, SVM, KNN | Sen = 95% Spe = 94.4% Acc = 94.7% (4th 1 min before) |
| Ebrahimzadeh et al. (2017) | 35 normal and 35 SCD (HRV) Source: NSR database and SCD database | 24 | Linear and non-linear methods (time-domain features (5); frequency-domain features (4); time-frequency domain features (11); non-linear features (4)) | MLP | Sen = 82.67% Spe = 85.09% Acc = 83.88% (12th 1 min before) |
| Ebrahimzadeh et al. (2018a) | 35 normal and 35 SCD (HRV) Source: NSR database and SCD database | 24 | Linear and non-linear methods (time-domain features (5); frequency-domain features (4); time -frequency domain features (11); non-linear features (4)) | MLP, KNN | Acc = 83.96% (4th 1 min before) |
| Ebrahimzadeh et al. (2019) | 35 normal and 35 SCD (HRV) Source: NSR database and SCD database | 28 | Linear and non-linear methods (time-domain features (5); frequency-domain features (4); time-frequency domain features (11); non-linear features (8)) | MLP, SVM, KNN | Sen = 85.72% Spe = 82.86% Acc = 84.28% (13th 1 min before) |
| Early SCD detection using HRV signals of 2-min interval | |||||
| Shen et al. (2007) | 20 normal and 23 SCD Source: NSR database and SCD database | 4 | Non-linear methods (time-frequency domain features (4)) | Artificial neural networks (ANN); back propagation (BP) | Acc = 87.5% (1st 2 min before) |
| Murukesan et al. (2014) | 18 normal and 20 SCD Source: NSR database and SCD database | 34 | Linear and non-linear methods, Poincaré plot analysis (time-domain features(15); frequency-domain features(13); non-linear features (6)) | SVM, PNN | Sen = 93.33% Spe = 100% Acc = 96.36% (1st 2 min before) |
| Current Study | 36 normal and 40 SCD Source: NSR database and SCD database | 27 | EEMD, linear and non-linear methods (time-domain features(3); frequency-domain features (4); non-linear features (5)) | KNN | Sen = 95%; Spe = 97.2% Acc = 96.1% (1st 2 min before) Average acc = 94.7% (14 min before) |
Sen, sensitivity; Spe, specificity; Acc, accuracy.