NSFA-Net

A noise-aware epileptic seizure prediction network that aligns multi-layer features and maintains the contextual consistency, achieving SOTA performance.

This project provides the code for model training and testing.

🚀 Quick Start

1. Prepare Dataset

We provide code for preprocessing the raw Kaggle dataset, which includes downsampling, segmentation, and STFT transformation. Use the following command to run the dataset preprocessing:

python Kaggle_preprocess_30.py --patient_id 1

Please download the raw Kaggle dataset via the following link:

Download Link for the raw Kaggle dataset

2. Run Training

The code for model training is located at train_NSFA.py. Use the following command to run the training for patient Dog_1:

python train_NSFA.py --patient_id 1 --device_number 0 --data_dir ../exp_30/Kaggle_dataset_30 --target_preictal_interval 30

Please replace the data_dir in the command with the actual path of your preprocessed data.

3. Run Evaluation

The code for model testing is located at test_Kaggle.py. Use the following command to run the evaluation for patient Dog_1:

python test_Kaggle.py --patient_id 1 --device_number 0 --batch_size 200 --ckpt_dir ../exp_30/NSFA_ckpt_30 --data_dir ../exp_30/Kaggle_dataset_30 --target_preictal_interval 30

Please replace the ckpt_dir and the data_dir in the command with the actual path of your model checkpoint and preprocessed data.