Dystonia is a severely disabling movement disorder with no cure, characterised by painful involuntary muscle contractions, twisting movements and abnormal postures. Causes include genetic conditions and brain injury, in particular dystonic cerebral palsy (CP) in which brain injury occurs around birth. Current therapies are limited: there is an urgent need to develop alternative, more personalised interventions. This requires greater understanding of pathophysiological mechanisms in dystonia.
Our group records electrical activity from brain (electroencephalography/EEG) and muscles (electromyography/EMG) and has demonstrated abnormal patterns of brain-muscle communication and cortical sensorimotor processing in childhood dystonia. In particular, modulation of the brain rhythm “mu”, which is typically suppressed in response to sensory stimulation or movement, is impaired in dystonia/dystonic CP, indicating abnormal sensorimotor processing.
We are currently investigating
a) whether children with dystonia/dystonic CP can enhance their mu modulation if provided with EEG-neurofeedback via a Brain-Computer-Interface;
b) development of mu modulation in healthy neonates and those at risk of developing dystonic CP.
Machine learning (ML) techniques offer a powerful approach to identify subtle patterns in complex EEG/EMG data.
Aims:
The student will apply ML approaches to investigate:
1. how deeper understanding of EEG/EMG signals and their interactions could refine childhood dystonia diagnosis and classification;
2. individualised responses to neurofeedback, potentially improving accessibility to Brain-Computer-Interface technology;
3. developmental patterns of mu modulation and their potential role in predicting outcomes.
Techniques/Skills:
Work with young people/patients and multi-disciplinary clinical teams;
Acquisition, analysis and preprocessing of EEG/EMG data from children/infants/toddlers;
Supervised and unsupervised ML methods;
Deep learning approaches, including Convolutional and Recurrent Neural Networks (CNNs/RNNs);
Python-based ML libraries (e.g., scikit-learn, TensorFlow, PyTorch)
Model validation, performance evaluation, interpretation of model outputs in relation to neurophysiological and clinical measures.
Statistical analysis, scientific writing, presentation skills.
Objectives:
Year_1: Training in EEG/EMG data collection/preprocessing, time and frequency domain analysis; Neonatal data collection;
Exploratory application in previously-acquired data of classical supervised learning methods (e.g., support vector machines, logistic regression, random forests) to classify dystonia phenotypes based on EEG/EMG features, and unsupervised learning techniques (e.g., k-means clustering, PCA) to identify potential subgroups based on sensorimotor signal profiles.
Year_2: Consolidation. Apply dimensionality reduction techniques to visualise complex EEG/EMG datasets. Deep learning models, e.g. CNNs to classify outcomes. Conference presentation.
Year_3: Refinement of deep learning architectures for modelling time-dependent features in neurofeedback and EEG data.
Year_4: Longitudinal ML modelling of neonatal EEG to capture developmental trajectories and predict later dystonia outcomes. Write-up. Conference presentation.
Rotation project:
Spectral analysis of EEG/EMG parameters