A transformer model for predicting cognitive impairment from sleep

Abstract

Sleep disturbances are known to be aggravated with normal aging. Additionally, sleep disruptions have a potentially bidirectional causal relationship with dementia due to neurodegenerative diseases like Alzheimer’s disease. Predictive techniques that can automatically detect cognitive impairment from an individual’s sleep data have broad clinical and biological significance. Here, we present a deep learning approach based on a transformer architecture to predict cognitive status from sleep electroencephalography (EEG) data. This work uses data from N = 1, 502 subjects from the Multi-Ethnic Study of Atherosclerosis (MESA) cohort. Our transformer model achieves 70.22% accuracy at the binary classification task for distinguishing cognitively normal and impaired subjects based on their sleep EEG. Our method outperforms traditional feature handcrafting, which has an overall accuracy of 57.61% for the same task. We use a sparse regression model to understand and interpret the information captured by each learned feature from our transformer model. To our knowledge, this is the first effort to use deep learning to predict cognitive impairment from sleep metrics.

Comprehensive review of Transformer‐based models in neuroscience, neurology, and psychiatry

“…Subject-dependent: 83.03% ACC Wei et al [187] TC-Net DEAP [188] EGG Emotion recognition Valence: 98.76% ACC Auxiliary emotion capsule module Arousal: 98.81% ACC Dominance: 98.82% ACC DREAMER [189] Valence: 98.59% ACC Arousal: 98.61% ACC Dominance: 98.67% ACC Sun et al [190] -DEAP [188] EGG Emotion recognition DEAP-arousal: 94.61% ACC Dual-branch dynamic graph convolution, transfer learning SEED [186] DEAP-valence: 95.91% ACC SEED-IV [191] SEED: 97.31% ACC SEED-IV: 89.97% ACC Xu et al [192] AMDET DEAP [188] EGG Emotion recognition DEAP-arousal: 97.48% ACC Interpretable model with Grad-CAM SEED [186] DEAP-valence: 96.85% ACC SEED-IV [191] SEED: 97.17% ACC SEED-IV: 87.32% ACC Wang et al [193] HSLT DEAP [188] EGG Emotion recognition DEAP-arousal: [194] 3DCGSA DEAP [188] EGG [196] -MESA [197] PSG Cognitive status prediction 70.22% ACC Predicting cognitive status from sleep metrics You et al [198] GTransU-CAP CAP sleep EGG A-phase detection Healthy: 67.78% F1 Gated Transformer-based U-Net Patients: 72.16% F1 Pradeepkumar et al [199] -Sleep-EDF-78 [200] EGG, fMRI indicating advancements in both methodology and conceptual comprehension within the field. Inspired by the ViT, Guo et al [206] introduced the deep convolutional Transformer (DCoT), which combines deep convolutional and Transformer encoders.…”

“…When evaluating their model using the Sleep Heart Health Study dataset, they achieved excellent classification results, with F1-scores of 0.92, 0.85, and 0.84 for the Wake, N2, and N3 stages, respectively. Recognizing the two-way relationship between sleep disorders and neurodegenerative diseases like AD, Song et al [196] proposed a deep-learning method based on the Transformer architecture for predicting cognitive states from sleep EEG data. Their model achieved an accuracy of 70.22% in a binary classification task.…”