Addressing data limitations in seizure prediction through transfer learning

Scripts and convolutional autoencoder model used for applying transfer learning in seizure prediction.

Authors

• Fábio Lopes
• Mauro F. Pinto
• António Dourado
• Andreas Schulze-Bonhage
• Matthias Dümpelmann
• César Teixeira

Files

• test_seizure_prediction_model_tl.py contains the code necessary to develop patient-specific models (it uses the seizure_prediction_model_cnn_lstm_autoencoder_128_last_approach.h5 as a transfer learning model). It also uses the statistics (average and standard deviation) from the dataset used to train the transfer learning model (standardisation_values_cnn_lstm_autoencoder_128_last_approach.npy).
• seizure_prediction_model_cnn_lstm_autoencoder_128_last_approach.h5 contains the weights of the transfer learning model in HDF5 file format.
• standardisation_values_cnn_lstm_autoencoder_128_last_approach.npy contains the average and standard deviation of the dataset used to train the transfer learning model.
• utils.py contains general functions used in the test_seizure_prediction_model_tl.py.

Requirements

• Python 3.7
• Tensorflow 2.6.0
• Numpy 1.19.5

Scientific Paper

• Lopes, F., Pinto, M.F., Dourado, A. et al. Addressing data limitations in seizure prediction through transfer learning. Sci Rep 14, 14169 (2024). https://doi.org/10.1038/s41598-024-64802-1