Explainable Machine-Learning-Based Characterization of Abnormal Cortical Activities for Working Memory of Restless...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging

Background:

• Restless legs syndrome (RLS) is a sensorimotor disorder characterized by an urge to move the legs and unpleasant leg sensations.
• RLS is frequently associated with prefrontal cortex dysfunction.
• Working memory deficits are commonly observed in individuals with RLS.

Purpose of the Study:

• To elucidate the neural mechanisms underlying working memory impairments in RLS.
• To apply machine learning to single-trial neural activity for RLS-related working memory analysis.

Main Methods:

• Development of a convolutional neural network classifier to distinguish cortical activity between RLS patients and controls.
• Utilizing layer-wise relevance propagation to identify critical neural activity patterns during working memory tasks.
• Analysis of single-trial event-related potentials with leave-one-subject-out cross-validation for robust classification.

Main Results:

• Achieved high classification accuracy (~94%) in differentiating RLS patients from controls using single-trial electroencephalography (EEG) data.
• Identified critical brain regions involved in working memory that show distinct activity patterns in RLS.
• Correlated neural activity in these critical areas with clinical RLS severity scores.

Conclusions:

• Machine learning effectively identifies neural markers of working memory deficits in RLS from noisy EEG data.
• The findings highlight the role of prefrontal cortex dysfunction in RLS-associated cognitive impairments.
• This approach offers a novel method for investigating the neural basis of sensorimotor disorders.