DiffECG: Diffusion Model-Powered Label-Efficient and Personalized Arrhythmia Diagnosis

To run the training and testing, you should download the dataset first by clicking here

After unzip the Data.zip, you wil get a Data fold with structure shown as follow:

├── Data
  ├── Dataset
  ├── Frequency
  └── Lorenz

You need to move all three subfolders to the root path of the project and you will get a file structure shown as follow:

├── Baselines
│  ├── MoCo
│  ├── EffNet
│  └── Models
├── Dataset
│  ├── data_ChapmanShaoxing_segments
│  └── data_LTAF_segments
├── Frequency
├── Lorenz
├── Diffusion_Based
├── requirements.txt
└── run.sh

The Baselines fold includes the pre-training, fine-tuning, and testing code of two baselines EfficientECG and MoCo.

The Diffusion_based fold includes the fine-tuning and testing code of the proposed diffusion-based method.

You can install the required package using the command pip install -r requirements.txt.

Pre-trained models are also provided, to conduct test with the pre-trained model, use following command:

bash ./run.sh <method> <task>

<method> can be selected in eff, moco, and diffEcg.

<task> can be selected in generalization and personalization.