This is a project which is currently making use of HPC facilities at Newcastle University. It is active.
For further information about this project, please contact:
This project aims to develop advanced deep learning frameworks for automated sleep stage classification using Polysomnography (PSG) signals. A critical challenge in this field is the high inter-subject variability which limits the generalization of current algorithms to new patients. We propose to research novel neural architectures that integrate spectral-spatial-sequential feature fusion and incorporate clinical contextual metadata to learn subject-invariant representations. By leveraging large-scale physiological datasets, the study seeks to bridge the gap between AI-based scoring and human clinical expertise, ensuring robust performance across diverse patient populations for assisting in sleep disorder diagnosis.
The research utilizes the Python ecosystem, specifically PyTorch, to implement complex deep learning models including Graph Neural Networks and State Space Models. We require High Performance Computing to process high-dimensional multi-channel time-series data (EEG, EOG, EMG). The computational workload involves training memory-intensive architectures and performing rigorous subject-wise cross-validation, which necessitates significant GPU memory and parallel processing capabilities to optimize composite loss functions and ensure model convergence.