Adewale, Q. | Search bioRxiv

bioRxiv2019

Mental Workload Estimation Using Wireless EEG Signals

Previous studies have shown that electroencephalogram (EEG) can be used in estimating mental workload. However, developing fast and reliable models for cross-task, cross-subject and cross-session classifications of workload remains a challenge. In this study, a wireless Emotiv EPOC headset was used to evaluate workload in two different mental tasks: n-back task and mental arithmetic task. 0-back task and 2-back task were employed as low and high workload in the n-back task while 1-digit and 3-digit addition were used as the two different workload levels in the arithmetic task. Using power spectral density as features, a fast signal processing and feature extraction framework was developed to facilitate real-time estimation of workload. Within-session accuracies of 98.5% and 95.5% were achieved in the n-back and arithmetic tasks respectively. Adaptive subspace feature matching (ASFM) was applied for cross-session, cross-task and cross-subject classifications. The feature adaptation provided average cross-session accuracies of 80.5% and 74.4% in the n-back and the arithmetic tasks respectively. An average cross-task accuracy of 68.6% was achieved while cross-subject accuracies were 74.4% and 64.1% in the n-back and arithmetic tasks respectively. The framework generalised well across subjects and tasks, and it provided a promising approach towards developing subject and task-independent models. This study also shows that a consumer-level wireless EEG headset can be applied in cognitive monitoring for real-time estimation of workload in practice.

neuroscience↗

bioRxiv2019

Blood and Brain Gene Expression Trajectories Underlying Neuropathology and Cognitive Impairment in Neurodegeneration

Neurodegenerative disorders take decades to develop and their early detection is challenged by confounding non-pathological aging processes. For all neurodegenerative conditions, we lack longitudinal gene expression (GE) data covering their large temporal evolution, which hinders the fully understanding of the underlying dynamic molecular mechanisms. Here, we aimed to overcome this limitation by introducing a novel GE contrastive trajectory inference (GE-cTI) method that reveals enriched temporal patterns in a diseased population. Evaluated on 1969 subjects in the spectrum of late-onset Alzheimers and Huntingtons diseases (from ROSMAP, HBTRC and ADNI studies), this unsupervised machine learning algorithm strongly predicts neuropathological severity (e.g. Braak, Amyloid and Vonsattel stages). Furthermore, when applied to in-vivo blood samples (ADNI), it predicts 97% of the variance in memory deterioration and its future declining rate, supporting the identification of a powerful and minimally invasive (blood-based) tool for early clinical screening and disease prevention. This technique also allows the discovery of genes and molecular pathways, in both peripheral and brain tissues, that are highly predictive of disease evolution. Eighty percent of the most predictive molecular pathways identified in the brain were also top predictors in the blood. The GE-cTI is a promising tool for revealing complex neuropathological mechanisms, with direct implications for implementing personalized dynamic treatments in neurology. HIGHLIGHTS- Unsupervised learning detects enriched gene expression (GE) trajectories in disease - These plasma and brain GE trajectories predict neuropathology and future cognitive impairment - Most predictive molecular functions/pathways in the brain are also top predictors in the plasma - By identifying plasma GE trajectories, patients can be easily screened and follow dynamic treatments

neuroscience↗