Table 3.
Literature review: XAI studies using ECG signals.
| References | Database | ECG input types | Target | Class | Deep method + explainability | Performance | |
|---|---|---|---|---|---|---|---|
| Signal | Cho et al.22 | Clinical data |
6-lead Signals 12-Lead Signals |
MI Detection | Normal and MI | CNN + GradCAM |
AUC: 0.880 AUC: 0.854 |
| Wang et al.16 |
MIT-BIH ADB PTB-XL DB |
Lead II Lead I, II, V1, V2, V3, V6 |
ECG Beat Classification | N, S, V, F, Q | ResNet + GradCAM | Acc: 98.64% | |
| Hicks et al.23 | Inter99 DB | 12-Lead Signals | Sex Prediction | Male, Female | CNN + GradCAM | Acc: 88.08% | |
| Ma et al.27 | MIMIC-III DB |
PPG Signals, Lead II Signals |
Blood Pressure Prediction | SBP and DBP | AlexNet, GoogLeNet, ResNet18, DenseNet121, DPN68 |
DenseNet121: MAE: 3.69, 1.95 RMSE: 5.18, 2.71 |
|
| Jahmunah et al.24 | PTB DB | 12-Lead Signals | MI Detection | Normal and 10 types of MI |
CNN + GradCAM DenseNet + GradCAM |
Acc: 98.9% | |
| Anand et al.25 |
PTB-XL DB Chapman |
12-Lead Signals |
- MI Detection - Arrhythmia Classification |
5 types of MI AFib, GSVT, SB, SR |
CNN + SHAP |
AUC: 93.41% AUC: 99.46% |
|
| Singh et al.32 | MIT-BIH ADB | Beat Signals | Arrhythmia Classification | 6 types of Arrhythmia |
CNN + ECANet + K-GradCAM |
Acc: 98.88% | |
| Kim et al.17 | Clinical data | Lead II Signals | Arrhythmia Detection | Normal and Arrhythmia | DenseNet + GradCAM | Acc: 98.71% | |
| Han et al.26 |
PTB DB PTB-XL DB |
12-Lead Signals | MI Detection and Localization | Normal and 5 types of MI | ResNet + GradCAM |
Acc: 99.97% Acc: 99.85% |
|
| Loh et al.33 | Clinical data | Segmented Raw ECG Signals | ADHD and CD Detection | ADHD, CD, ADHD + CD | CNN + GradCAM | Acc: 96.04% | |
| Le et al.19 | Chapman, CPSC | Lead I, II, V1 ECG Signals | Arrhythmia Classification | 13 types of Arrhythmia | 1DSEResNet + GradCAM |
F1 Score: 0.9718 F1 Score: 0.8004 |
|
| Image | Makimoto et al.31 | PTB DB | 9-Lead PNG formatted Images | MI Detection | Normal and MI | CNN + GradCAM | Acc: 87.5% ± 11.8% |
| Sangha et al.28 |
CODE study PTB-XL DB |
12-Lead PNG formatted Images | Multilabel Prediction | Male and 6 types of Arrhythmia | EfficientNet + GradCAM |
AUC: 0.99 AUC: 0.97 |
|
| Fang et al.29 |
PTB DB PTB-XL DB |
12-Lead 3-D ECG Images | MI Detection | Normal and MI | CNN + GradCAM |
Acc: 95.65% Acc: 97.23% |
|
| Varandas et al.20 | MIT-BIH ADB | Lead II, V5 ECG Images | Arrhythmia Detection | Normal and Arrhythmia | ResNet50 + GB GradCAM |
Acc: 93.66% Acc: 91.72% |
|
| Kondo et al.18 | Clinical data |
Lead II, V3, V5, aVR ECG Images |
CCU Mortality Prediction | CCU Survivors and Non-survivor |
CNN + GradCAM |
Acc: 77.3% | |
| Ao et al.30 |
PTB-XL DB, CPSC, Chapman, Tongji |
12-Lead ECG Images | Arrhythmia Classification | Diagnoses of 15 types | VGG16 + GradCAM | AUC: 0.982 | |
| Image + Signal | Our study | Chapman21 | Lead II | Arrhythmia Classification | AFib, GSVT, SB, SR | CNN + MM-GradCAM |
Signal Form: Acc: 93.07% Image Form: Acc: 97.44% |
Acc: Accuracy, Sen: Sensitivity, Spec: Specificity, AUC: area under the receiver operating characteristics curve, N: non-ectopic, S: supraventricular ectopic, V: ventricular ectopic, F: Fusion, Q: Unknown, SBP: Systolic Blood Pressure, DBP: Diastolic Blood Pressure, MAE: Mean Absolute Error, RMSE: Root Mean Square Error, ADHD: Attention Deficit Hyperactivity Disorder, CD: Conduct Disorder, CCU: Cardiac Care Unit.