A Few-Shot Learning-Based EEG and Stage Transition Sequence Generator for Improving Sleep Staging Performance.

In this study, generative adversarial networks named SleepGAN are proposed to expand the training set for automatic sleep stage classification tasks by generating both electroencephalogram (EEG) epochs and sequence relationships of sleep stages. In order to reach high accuracy, most existing classification methods require substantial amounts of training data, but obtaining such quantities of real EEG epochs is expensive and time-consuming. We introduce few-shot learning, which is a method of training a GAN using a very small set of training data. This paper presents progressive Wasserstein divergence generative adversarial networks (GANs) and a relational memory generator to generate EEG epochs and stage transition sequences, respectively. For the evaluation of our generated data, we use single-channel EEGs from the public dataset Sleep-EDF. The addition of our augmented data and sequence to the training set was shown to improve the performance of the classification model. The accuracy of the model increased by approximately 1% after incorporating generated EEG epochs. Adding both the augmented data and sequence to the training set resulted in a further increase of 3%, from the original accuracy of 79.40% to 83.06%. The result proves that SleepGAN is a set of GANs capable of generating realistic EEG epochs and transition sequences under the condition of insufficient training data and can be used to enlarge the training dataset and improve the performance of sleep stage classification models in clinical practice.

Automatic sleep stage classification: A light and efficient deep neural network model based on time, frequency and fractional Fourier transform domain features.

This work proposed a novel method for automatic sleep stage classification based on the time, frequency, and fractional Fourier transform (FRFT) domain features extracted from a single-channel electroencephalogram (EEG). Bidirectional long short-term memory was applied to the proposed model to train it to learn the sleep stage transition rules according to the American Academy of Sleep Medicine's manual for automatic sleep stage classification. Results indicated that the features extracted from the fractional Fourier-transformed single-channel EEG may improve the performance of sleep stage classification. For the Fpz-Cz EEG of Sleep-EDF with 30 s epochs, the overall accuracy of the model increased by circa 1% with the help of the FRFT domain features and even reached 81.6%. This work thus made the application of FRFT to automatic sleep stage classification possible. The parameters of the proposed model measured 0.31 MB, which are 5% of those of DeepSleepNet, but its performance is similar to that of DeepSleepNet. Hence, the proposed model is a light and efficient model based on deep neural networks, which also has a prospect for on-device machine learning.