Preprocessed CPSC and PTB-XL Data

CPSC 2018The first dataset is a preprocessed version of the CPSC 2018 dataset, which contains 6877 ECG recordings. We preprocessed the dataset by resampling the ECG signals to 250 Hz and equalizing the ECG signal length to 60 seconds, yielding a signal length of T=15,000 data points per recording.For the hyperparameter study, we employed a fixed train-valid-test split with ratio 60-20-20, while for the final evaluations, including the comparison with the state-of-the-art methods and ablation studies, we used a 10-fold cross-validation strategy.The raw CPSC 2018 dataset can be downloaded from the website of the PhysioNet/Computing in Cardiology Challenge 2020.(License: Creative Commons Attribution 4.0 International Public License).PTB-XL (Super-Diag.)The second dataset is a pre-processed version of PTB-XL, a large multi-label dataset of 21,799 clinical 12-lead ECG records of 10 seconds each. PTB-XL contains 71 ECG statements, categorized into 44 diagnostic, 19 form, and 12 rhythmic classes. In addition, the diagnostic category can be divided into 24 sub- and 5 coarse-grained super-classes. In our pre-processed version, we utilize the super-diagnostic labels for classification and the recommended train-valid-test splits, sampled at 100 Hz. We select only samples with at least one label in the super-diagnostic category,without applying any further preprocessing.The raw PTB-XL dataset can be downloaded from the PhysioNet/PTB-XL website.(License: Creative Commons Attribution 4.0 International Public License).