A 300-second example of a 13-channel physiological recording and the corresponding sleep arousal prediction/target labels.

DeepSleep 2.0 is a compact version of DeepSleep, a state-of-the-art, U-Net-inspired, fully convolutional deep neural network, which achieved the highest unofficial score in the 2018 PhysioNet Computing Challenge. The proposed network architecture has a compact encoder/decoder structure containing only 740,551 trainable parameters. The input to the network is a full-length multi-channel polysomnographic recording signal. The network has been designed and optimized to efficiently predict non-apnea sleep arousals on held-out test data at a 5-millisecond resolution level, while not compromising the prediction accuracy. When compared to DeepSleep, the obtained experimental results in terms of gross area under the precision-recall curve (AUPRC) and gross area under the receiver operating characteristic curve (AUROC) suggest that a lightweight architecture, which can achieve similar prediction performance at a lower computational cost, is realizable.

It is assumed that you have the full or partial PhysioNet dataset (~135 GB of data per folder) on the disk. In ./data, you can find two bash scripts to download the PhysioNet dataset.

Here are the essential steps to sucesfully run the main Jupyter notebook file (deep_sleep2.ipynb).

STEP 0: Clone the Repository

STEP 1: Installation

STEP 2: Hyperparameters

A correctly set up hyperparameters.txt file must be present in a subdirectory of ./models. The subdirectory name is specified in the MODEL_NAME variable.

STEP 3: Notebook File

Run the cells of deep_sleep2.ipynb in a sequential order. Consider the description of the Main Switches section.

If you use this code in your research, please cite the following publication:

Consider also citing the original DeepSleep paper.