Unfolding the Effects of Acute Cardiovascular Exercise on Neural Correlates of Motor Learning Using Convolutional Neural Networks

Frontiers in Neuroscience

Marc Roig¹  Marie-Hélène Boudrias¹  Georgios D. Mitsis²  Arna Ghosh³  Fabien Dal Maso⁵ 

[1] Center for Interdisciplinary Research in Rehabilitation of Greater Montreal (CRIR), Montreal, QC, Canada;Department of Bioengineering, McGill University, Montreal, QC, Canada;Integrated Program in Neuroscience, McGill University, Montreal, QC, Canada;School of Physical and Occupational Therapy, McGill University, Montreal, QC, Canada;École de Kinéiologie et des Sciences de l'Activité Physique, Université de Montréal, Montreal, QC, Canada;

关键词: motor learning;    convolutional neural network (CNN);    cardiovascular exercise;    deep learning;    EEG;   

DOI  :  10.3389/fnins.2019.01215

来源: DOAJ

【 摘 要 】

Cardiovascular exercise is known to promote the consolidation of newly acquired motor skills. Previous studies seeking to understand the neural correlates underlying motor memory consolidation that is modulated by exercise, have relied so far on using traditional statistical approaches for a priori selected features from neuroimaging data, including EEG. With recent advances in machine learning, data-driven techniques such as deep learning have shown great potential for EEG data decoding for brain-computer interfaces, but have not been explored in the context of exercise. Here, we present a novel Convolutional Neural Network (CNN)-based pipeline for analysis of EEG data to study the brain areas and spectral EEG measures modulated by exercise. To the best of our knowledge, this work is the first one to demonstrate the ability of CNNs to be trained in a limited sample size setting. Our approach revealed discriminative spectral features within a refined frequency band (27–29 Hz) as compared to the wider beta bandwidth (15–30 Hz), which is commonly used in data analyses, as well as corresponding brain regions that were modulated by exercise. These results indicate the presence of finer EEG spectral features that could have been overlooked using conventional hypothesis-driven statistical approaches. Our study thus demonstrates the feasibility of using deep network architectures for neuroimaging analysis, even in small-scale studies, to identify robust brain biomarkers and investigate neuroscience-based questions.

【 授权许可】

Unknown