Table 8 Summary of performance comparison for different methods
From: Heart failure classification using deep learning to extract spatiotemporal features from ECG
Classification problem | Model | Number of data | Performance |
|---|---|---|---|
Two classes | SVM-GA [14] | Clinical Data NYHA class III: 1365 NYHA class IV: 2522 | Acc – 91.49% Ppv – 94.25% Recall–93.60% |
11-layer CNN [31] | 5-seconds ECG segment CHF: 30000 Normal: 70308 | Acc – 98.97% Sen – 98.87% Spe – 99.01% | |
Three classes | CART [10] | RR interval segment (N = 300) NYHA class I: 1416 NYHA class II: 3088 NYHA class III: 6181 | Acc – 81.40% Sen – 66.50% Spe – 81.60% |
AdaBoost [13] | Poincaré plot NYHA class I: 22 NYHA class II: 116 NYHA class III: 61 | Acc – 82.5% Ppv – 77.8% Sen – 58.3% Spe – 92.9% | |
Four classes | NLP [9] | Clinical note NYHA class I: 1367 NYHA class II: 2502 NYHA class III: 1790 NYHA class IV: 515 | Ppv – 94.99% Recall–92.10% |
Multi-scale ResNet-34[12] | 5-seconds ECG segment NYHA class I: 3720 NYHA class II: 7440 NYHA class III: 11940 NYHA class IV: 6240 | Acc – 94.29% Ppv – 94.16% Sen – 93.79% Spe – 97.89% | |
Our work | 12-seconds ECG segment NYHA class I: 1200 NYHA class II: 7050 NYHA class III: 17250 NYHA class IV: 14700 | Acc – 99.09% Ppv – 98.98% Sen – 99.03% Spe – 99.64% | |
Five classes | CNN-RNN [11] | 2-seconds ECG segment Normal: 5160 NYHA class I: 2520 NYHA class II: 4680 NYHA class III: 3150 NYHA class IV: 6240 | Acc – 97.60% Ppv – 97.10% Sen – 96.30% Spe – 97.40% |