Rank,Paper Title,Score,Data preprocessing,Data augmentation,Imbalance,End-to-end,Ensemble,Multi-binary classification,Feature engineering,Hand Features,Temporal,Frequent,Demographic Features,DNN,CNN,RNN/Transformer,Attention,Backbone,Others
1,A Wide & Deep Transformer Neural Network for 12-Lead ECG Classification,0.533,Sampling rate 500Hz; Bandwidth 3 - 45 Hz; Value -1 - 1; Window size 15s 0-padding,No,No,Yes,Top 3 best models from 10 folds CV,Yes,Yes,No,No,No,"age,sex + 20 top identified features by random forest model",Yes,Yes,Transformer,Inherently in Transformer (location wise),ResNet + Transformer + hand features + ensemble,None
2,Adaptive lead weighted ResNet trained with different duration signals for classifying 12-lead ECGs,0.52,Sampling rate 257 Hz; Value -1 - 1; Window size 4096(16s) 0-padding,No,Thresholding,Yes,Ensemble 5 CV models,Yes,Yes,No,No,No,only age and sex,Yes,Yes,No,Squeeze-And-Excitation (channel wise),ResNet + SE,None
3,Classification of Cardiac Abnormalities from ECG Signals using SE-ResNet,0.514,Sampling rate 500Hz; Window size 30s 0-padding; Exclude 4 leads; Wavelet denoise,With external data: Hefei,Sign loss,Yes,Rule model for bradycardia,Yes,No,No,No,No,No,Yes,Yes,No,Squeeze-And-Excitation (channel wise),ResNet + SE,None
4,Combining Scatter Transform and Deep Neural Networks for Multilabel Electrocardiogram Signal Classification,0.485,Sampling rate 500Hz; Value -1 - 1; Window size 10240(20.48 s) ,Add noise and drift,No,Yes,No,Yes,No,No,No,No,No,Yes,Yes,Transformer,Inherently in Transformer (location wise),ResNet + Transformer,None
5,Classification of 12-lead ECG Signals with Adversarial Multi-Source Domain Generalization,0.437,Sampling rate 250Hz; Value normalized,"Add or filter frequency components; Substitute, shuffle, invert, filt and scale lead data",No,Yes,No,Yes,Yes,No,No,No,only age and sex,Yes,Yes,Yes,No,ResNet + RNN,Domain-adversarial training
6,Bag of Tricks for Electrocardiogram Classification with Deep Neural Networks,0.42,Sampling rate 500Hz; Bandwith 50-60Hz; Value use mean and std; Exlude PTB and StPetersburg datasets,Manifold Mixup,"Weighted loss, Thresholding",Yes,Ensemble 10 CV models,Yes,No,No,No,No,No,Yes,Yes,No,No,ResNet,None
7,Automated Comprehensive Interpretation of 12-lead Electrocardiograms Using Pre-trained Exponentially Dilated Causal Convolutional Neural Networks,0.417,Sampling rate 500Hz,With external data: UMCU,Weighted loss (focal),Yes,No,Yes,No,No,No,No,No,Yes,Yes,No,No,ResNet,None
8,SE-ECGNet : Multi-scale SE-Net for Multi-lead ECG Data,0.411,Window sizw 10s,Random shift signals; Add Gaussian noise,Thresholding,Yes,No,Yes,Yes,No,No,No,only age and sex,Yes,Yes,No,Squeeze-And-Excitation (channel wise),ResNet + SE,None
9,Impact of Neural Architecture Design on Cardiac Abnormality Classification Using 12-lead ECG Signals,0.382,Window size 10s 0-padding,No,Thresholding,Yes,No,Yes,No,No,No,No,No,Yes,Yes,Yes,No,ResNet + BiLSTM,NAS
10,Automatic Detection and Classification of 12-lead ECGs Using a Deep Neural Network,0.359,Sampling rate 500Hz; Window size 10s 0-padding,No,No,Yes,No,Yes,Yes,No,No,No,only age and sex,Yes,Yes,No,Squeeze-And-Excitation (channel wise),ResNet + SE,None
11,Cardiac Pathologies Detection and Classification in 12-lead ECG,0.354,No,No,No,No,No,Yes,Yes,Yes,Yes,Yes,No,No,No,No,No,Rule based,None
12,Rule-Based Method and Deep Learning Networks for Automatic Classification of ECG,0.298,Sampling rate 500 Hz,No,"At most N_max
instances",No,No,Yes,Yes,Yes,Yes,Yes,No,Yes,Yes,No,No,Rule,CWT transforms signal to scalogram image
13,Convolutional Recurrent Neural Network and LightGBM Ensemble Model for 12-lead ECG Classification,0.281,Samping rate 500Hz; Bandwidth 0.01 - 20 Hz; Window size 10s ,No,SMOTE,No,Yes,Yes,Yes,Yes,Yes,Yes,only age and sex,Yes,Yes,Yes,Yes,CRNN + LightGBM,None
14,12-lead ECG Arrythmia Classification Using Convolutional Neural Network for Mutually Non-Exclusive Classes,0.279,"Samping rate 100Hz, Window size 10 s; Median filter with 5 samples;  Subtract mean of 50 samples",No,No,Yes,No,Yes,No,No,No,No,No,Yes,Yes,No,No,CNN,None
15,Cardiac Abnormality Detection in 12-lead ECGs With Deep Convolutional Neural Networks Using Data Augmentation,0.27,Sampling rate 500 Hz; Bandwidth 0.5 - 45 Hz; Uniform filter with 50 samples,Signal slicing; CutOut; Channel shifting; Additive noise; Dropout,No,Yes,No,No,No,No,No,NO,NO,Yes,Yes,No,No,ResNet,Data augmentation
16,A Bio-toolkit for Multi-Cardiac Abnormality Diagnosis Using ECG Signal and Deep Learning,0.26,Cardiac cycle 400 samples; Bandwidth 3 - 45 Hz;,No,No,Yes,No,No,No,No,No,No,No,Yes,Yes,LSTM,No,CNN+LSTM,None
17,Multi-label Arrhythmia Classification From 12-Lead Electrocardiograms,0.24,Bandwidth 12 - 50 Hz; Median filter with 10ms; ,No,Undersample the majority class,No (Feature extraction),No,Yes,Yes,Yes,Yes,Yes,No,Yes,Yes,Yes,No,"AlexNet, LSTM, SVM… for different classes",Feature Map Generation algorithm design
18,MADNN : A Multi-scale Attention Deep Neural Network for Arrhythmia Classification Data preprocessing,0.24,"Sampling rate 500 Hz, Window size 30000(60 s) copy-padding",Romdomly crop,Focal loss; Balance factor to filter samples,Yes,"VGG, Xception",No,No,No,No,No,No,Yes,Yes,No,Yes,ResNet+SE+SK,Data augmentation
19,Classification of 12-lead ECGs Using Gradient Boosting on Features Acquired With Domain-Specific and Domain-Agnostic Methods,0.23,Bandwidth 0-80 Hz; Median filter with 200 ms and 600 ms,No,Re-weight positive samples,No (Feature extraction),No,No,Yes,Yes,Yes,Yes,No,No,No,No,No,XGBoost,Feature extraction
20,Convolutional Neural Network and Rule-Based Algorithms for Classifying 12-lead ECGs,0.21,Samping rate 500Hz; Window size 5000(10s) 0-padding,No,No,Yes,No,No,Yes,No,No,No,only age and sex,Yes,Yes,No,No,CNN+Rule-based model (HRV-score),RMSSD
21,Automated Classification of Electrocardiograms Using Wavelet Analysis and Deep Learning,0.21,No,No,Undersample the majority class,No,No,Yes,Yes,Yes,No,Yes,No,Yes,Yes,No,No,"SqueezeNet, 12 models for each lead with voting scheme",None
22,ECG Abnormalities Recognition Using Convolutional Network With Global Skip Connections and Custom Loss Function,0.2,Sampling rate 125 Hz,No,WCE loss function,Yes,No,No,No,No,No,No,No,Yes,Yes,No,No,ResNet+Custom Loss Function,None
23,Multi-Class Classification of Pathologies Found on Short ECG Signals,0.19,Samping rate250Hz,No,No,Yes,No,Yes,Yes,Yes,Yes,No,No,No,No,No,No,"ECOC, Bagging",None
24,ECG Classification with a Convolutional Recurrent Neural Network,0.17,"Samping rate 257 Hz, Bandwidth 0.5- Hz",Random offset,Focal loss; Oversampling; MLSMOTE;  An ensemble of binary and multi-label deep models,Yes,No,No,No,No,No,No,No,Yes,Yes,GRU,No,CRNN,Time Augmentation
25,Interpretable XGBoost Based Classification of 12-lead ECGs Applying Information Theory Measures From Neuroscience,0.16,Bandwidth 1.5-25 Hz; db4,No,Undersample the majority class,No (Feature extraction),No,Yes,Yes,Yes,Yes,Yes,known patient metadata.,No,No,No,No,XGBoost,None
26,Identification of Cardiac Arrhythmias from 12-lead ECG using Beat-wise Analysis and a Combination of CNN and LSTM,0.15,"Sampling rate 500 Hz; Window size 10s 0-padding, Bandwith 0-50 Hz; MODWT; db5; LOESS",No,Select most significant beats number,Yes,No,No,No,No,No,No,No,Yes,Yes,LSTM,No,CNN+BiLSTM,None
27,Classification of 12-Lead Electrocardiograms Using Residual Neural Networks and Transfer Learning,0.14,Sampling rate 250Hz; Window size 10s,With external data: MUSE,No,Yes,No,No,No,No,No,No,No,Yes,Yes,No,No,ResNet,None
28,ECG Segmentation Using a Neural Network as the Basis for Detection of Cardiac Pathologies,0.14,Sampling rate 500 Hz; Bandwith 0.05-42 Hz; db4,No,Focal loss,No (Feature extraction),No,Yes,Yes,Yes,Yes,No,No,Yes,Yes,No,No,Unet+XGBoost,None
29,Automatic 12-lead ECG Classification Using a Convolutional Network Ensemble,0.132,Sampling rate 400hz; Window size 4096 0-padding,No,Correlation factor,Yes,7 convolutional models ,No,No,No,No,No,No,Yes,Yes,No,No,ResNet,None
30,Utilization of Residual CNN-GRU With Attention Mechanism for Classification of 12-lead ECG,0.122,Sampling rate 250 Hz,Power envelopes,No,Yes,Yes,No,No,No,No,No,No,Yes,Yes,GRU,No,ResNet + RNN,None
31,Selected Features for Classification of 12-lead ECGs,0.102,Bandwidth 1- Hz; Median filter ,No,No,No,Bootstrap-aggregated ,No,Yes,Yes,Yes,No,only age and sex,No,No,No,No,Decision tree,Featre extraction
32,Classification of 12 Lead ECG Signal Using 1D-Convolutional Neural Network With Class Dependent Threshold,0.077,Sampling rate 500 Hz; Window size 10000 copy-padding; Converted to physical unit(mV),No,No,Yes,No,Yes,No,No,No,No,No,Yes,Yes,No,No,CNN,None
33,Deep Multi-Label Multi-Instance Classification on 12-Lead ECG,0.001,Bandwidth 0.5 - 50 Hz,Randomly sampled,BCELoss,Yes,No,Yes,No,No,No,No,No,Yes,Yes,No,Yes,ResNet Encoder+attention-based MIC+Decoder,None
34,Diagnostic of Multiple Cardiac Disorders from 12-lead ECGs Using Graph Convolutional Network Based Multi-label Classification,-0.012,Sampling rate 500 Hz; Window size 18s 0-padding,No,No,Yes,Yes,Yes,No,No,No,No,No,Yes,Yes,GRU,No,CNN+BiGRU+GCN,CNN+BiGRU+GCN(GRN for label correlations)
35,Automatic Concurrent Arrhythmia Classification Using Deep Residual Neural Networks,-0.035,Window size 2500,No,No,Yes,Yes,No,No,No,No,No,No,Yes,Yes,RNN GRU LSTM,Yes,ResNet,None
36,Multilabel 12-Lead Electrocardiogram Classification Using Gradient Boosting Tree Ensemble,-0.08,"Sampling rate 0-500 Hz; Window middle 2,000 samples(R:-0.35-0.5s); Average filter with 0.02s; ",No,Oversampling; Re-weight,No,Yes,Yes,Yes,Yes,Yes,No,No,No,No,No,No,XGboost,None
37,A Topology Informed Random Forest Classifier for ECG Classification,-0.113,1800 time points,No,No,No,No,No,Yes,Yes,Yes,No,demographic data,No,No,No,No,random forest,point cloud embedding
38,Detection of Cardiac Arrhythmias From Varied Length Multichannel Electrocardiogram Recordings Using Deep Convolutional Neural Networks,-0.128,Window size 3000 0-padding,No,Thresholding,Yes,No,No,No,No,No,No,No,Yes,Yes,No,No,VGG16,None
39,Multi-Stream Deep Neural Network For 12-Lead ECG Classification,-0.29,Bandwidth 0.1-30 Hz ,No,No,No (Feature extraction),Yes,No,Yes,Yes,Yes,No,No,Yes,Yes,No,No,CNN,CNN+feature extraction
40,Classification of 12-lead ECG with an Ensemble Machine Learning Approach,-0.476,Sampling rate 500 Hz; Bandwidth 0.67-30 Hz,Vector magnitude for detect beats,No,No (Feature extraction),Yes,Yes,Yes,Yes,Yes,No,No,Yes,Yes,No,No,CNN,3 CNNs for signal + FFNN for phythm + PVC detection
41,Classification of 12-lead ECGs using digital biomarkersand representation learning,-0.658,Sampling rate 500 Hz;  Bandwidth 0.13-88 Hz; bSQ>0.8;,No,Oversampling,No,Yes,No,Yes,Yes,Yes,No,only age and sex,Yes,Yes,GRU,No,CNN+GRU,feature engineering + CNN