Computational Framework Enabling an EEG-based BCI for Neurofeedback in Language Disorders: The case of dyslexia

Neurofeedback has shown promising results in ADHD, stroke and traumatic brain injury patients but has not previously been used to remediate the cognitive symptoms of developmental language disorders. One major limitation to the advancement of this field has been the absence of a framework combining the Brain-Computer Interface (BCI) literature with the developmental neurobiology of speech and language. This thesis establishes one possible theoretical, technical, and scientific basis for an operant-learning EEG-BCI targeted at developmental language disorders. Here, the chosen developmental language disorder of focus was dyslexia, due to the availability in-house of pre-existing EEG datasets. The BCI targets the cognitive symptoms of developmental dyslexia, presenting an initial set of experimental results with an adult population. First, statistical modelling was applied to the child EEG datasets to identify neural patterns associated with atypical speech processing and dyslexia in children. This statistical modelling was informed by the neuroscience of speech literature and the temporal sampling theory of dyslexia and two potential neural patterns to target for neurofeedback were identified. The first was an hypothesis-driven theta/delta oscillatory band ratio observable in a story listening condition (connected speech). The second was a linear classifier which predicted dyslexia status (0.77 AUC). This classifier was based on supervised spatial filtering of delta-band oscillations, showing useful features to classify dyslexic neural patterns both across connected speech and rhythmic syllable repetition tasks. Next, a BCI was designed to allow participants to learn how to self-regulate the first of these neural patterns via operant learning. The design of the BCI took into account the need for BCI control to be adapted to each individual’s performance style, without changing the core neural pattern being targeted. The BCI was played by 15 adults (7 dyslexics) who performed an EEG-BCI neurofeedback protocol of 16 sessions administered across two weeks. The majority of these adults significantly improved their BCI performance. BCI training had some effects on subsequent phonological processing for the adults with dyslexia. Potential generalizations of this study and issues regarding future BCI design, including the adaptation of this neurofeedback protocol for randomized control trials, are discussed.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2023

Author dc:creator

• Fonseca de Araujo, Joao

Advisor dc:contributor.advisor

• Goswami, Usha

• Brain-Computer Interfaces
• Common Spatial Patterns
• Developmental Language Disorders
• Digital Signal Processing
• Dyslexia
• EEG
• Machine Learning
• Speech Neuroscience

Licence

• https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/733cb97d-c9d7-43b3-8974-bd9aa4923b3b/download
• https://creativecommons.org/licenses/by/4.0/

Language dc:language

eng

DOI dc:identifier.doi

https://doi.org/10.17863/CAM.109624

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/370040