A Flexible Deep Learning Architecture for Temporal Sleep Stage Classification using Accelerometry and Photoplethysmography[^1]

[Paper|Presentation]

Conceptual representation of the proposed deep neural network (DNN) in an example recording. Two time-aligned spectrograms are firstly concatenated, reshaped, and zero-padded to conform to the subsequent temporal module. Then the segments are processed in the deep convolutional neural network, inspired by U-Net[^2][^3][^4] that consists of 𝑀 encoder and decoder blocks. Finally, the output is segmented into sleep epochs of 30 s duration and classified into 4 classes: wake, light sleep, deep sleep. The classification module is inspired by the segment classifier from U-Sleep[^3]. The argmax of the model predictions is presented along with the ground truth hypnogram for comparison. Periods with data loss are labeled with mask. GELU: Gaussian Error Linear Unit activation function; conv: convolution; convTranspose: transposed convolutional; batch norm: batch normalization; STFT: Short Time Fourier Transform; ACC: Accelerometry; PPG: Photoplethysmography.

[^1]: M. Olsen, J. M. Zeitzer, R. N. Richardson, P. Davidenko, P. J. Jennum, H. B. D. Sørensen, and E. Mignot. "A flexible deep learning architecture for temporal sleep stage classification using accelerometry and photoplethysmography," IEEE Transactions on Biomedical Engineering, 2022.

[^2]: O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 9351, pp. 234–241, 2015.

[^3]: M. Perslev, S. Darkner, L. Kempfner, M. Nikolic, P. J. Jennum, and C. Igel, “U-Sleep: resilient high-frequency sleep staging,” npj Digit. Med., vol. 4, no. 1, pp. 1–12, 2021.

[^4]: H. Li and Y. Guan, “DeepSleep convolutional neural network allows accurate and fast detection of sleep arousal,” Commun. Biol., vol. 4, no. 1, pp. 1–11, 2021.

Minimal example

Use minimal_example to get started.

Data availability

The authors do not have permission to share the data used in this study at this point in time. However, the pretrained model is available. Use minimal_example to get started.

Reproduction of all experiments in paper

Updates will be coming soon.