This repo is the official implementation of "Using Multi-Task Learning-based Framework to Detect ST-segment and J-point Deviation from Holter".
This including the whole train code of our proposed transformed based model.
This two Holter ECG datasets (R-ECG and E-ECG) from Ruijing Hospital collected for this study will be available after approval by the corresponding author.
We proposed a Transformer based structure to precisely detect the location and deviation of ST-segment and J point on 12 leads Holter ECG data in a beat level and provide cardiologists more accurate information about myocardial ischemia.
- To train denoise model, manual generate train and val csv in following path
It should only contain normal ECG feather paths, The noise csv is provied by this this repo:
./data/noise/train_denoise.csv
./data/noise/val_denoise.csv
the denoise ECG data is saving in feather, each feather file contains 12 leads.
the example feather files are proven in ./data/noise/
- To train segmentation model, manual generate train and val csv in following path
the example csv is provided, contain the data path.
./data/segmentation/train_segmentation.csv
./data/segmentation/val_segmentation.csv
the segmentation ECG data is saving in csv, each feather file contains 12 leads.
the example feather files are proven in ./data/segmentation/
the model is training with config file, which setting the dataset and hypermeter once set the config,
to train denoise
python3 train_unet_denoise.py --config ./configs/denoise.yaml
to train segmentation
python3 train_unet_semseg.py --config ./configs/segmentation.yaml