Deep Learning for Magnetoencephalography
Deep learning has been used in a wide range of applications, but it has only very recently been applied to Magnetoencephalography which is used to understand a variety of cognitive processes; for instance it can be used to understand how we process language or identify cognitive decline such as dementia. Work published in 2019 showed that it was possible to apply deep learning to categorise induced responses to stimuli across subjects. While trailblazing in its application of deep learning, it used relatively simple neural network (NN) models compared to other domains such as image and natural language processing.
In these other domains, there is a long history in developing complex NN models that combine spatial and temporal information in a range of ways. This thesis proposes more complex NN models that focus on modeling temporal relationships in the data, and applies them to the challenges of MEG data such as vulnerability to noise. In addition, it explores other insights from image processing to this domain, such as the unexpectedly high importance of approaches to data normalization. It applies these techniques to an extended range of MEG-based tasks, and finds that our new NN models outperform existing work on temporally-oriented tasks.