ABSTRACT
Rapid eyeball movement (REM) based sleep behavior disorder (SBD), or REM-SBD, is a relatively rare neurodegenerative disorder affecting more than 1% of people around the world. In this fatal condition, the brain remains at a conscious state but the body is temporarily paralyzed, counting 25% of total sleep quality. Traditional techniques used for tracking the sleep pattern of this disorder through statistical classifiers are too tedious, inconvenient, and subjective to identify the patterns with significant specificity and sensitivity. Therefore, this chapter targets recognizing dynamic sleeping patterns of REM-SBD using artificial intelligence. The proposed work starts with collecting a polysomnographic database of 16 patients with sleep-scoring sub-band EEG epochs. Later on, features are extracted by using the wavelet packet decomposition (WPD) method, and the extracted features are fed to two different neural network-based classifiers such as deep neural network (DNN) for classification and long term short memory (LTSM). It is evident from the proposed work that DNN used in compressing the lower weighted features gives better performance with the raw database and requires a larger training dataset. However, LTSM extracts more features and automatically adjusts the gradients, and it showed a better rate of convergence than DNN.
