• Huy Phan, Kaare Mikkelsen, Oliver Y. Chén, Philipp Koch, Alfred Mertins, and Maarten De Vos. SleepTransformer: Automatic Sleep Staging With Interpretability and Uncertainty Quantification. IEEE Transactions on Biomedical Engineering (TBME), vol. 69, no. 8, pp. 2456-2467, 2022. [PDF] [Preprint]
  

These are source code and experimental setup for SHHS.

1. Download the database SHHS. This may require to apply for licences. Information on how to obtain it can be found in the corresponding website.
2. To prepare data run shhs_data.m
3. Split data and generate file lists run data_split_eval.m run genlist_scratch_training.m (Not: I have included the "data_split_eval.mat" file and the "file_list" folder, you dont have to run this step again)
4. Training and evaluation run bash scripts in "scratch_training/sleeptransformer". The environment I used was Tensorflow 1.13, Python 3.7
5. Run matlab scripts in "evaluation" folders to aggregate the network outputs and compute metrics for example, run aggregate_sleeptransformer.m

The model weights trained on SHHS are available at https://zenodo.org/record/7927282 for reproducing the results in the paper.

• Matlab v7.3 (for data preparation)
• Python3.7
• Tensorflow GPU 1.x (x >= 3) (for network training and evaluation)
• numpy
• scipy
• h5py
• sklearn

CC-BY-NC-4.0