Technical Program

Paper Detail

Paper: PS-1A.31
Session: Poster Session 1A
Location: Symphony/Overture
Session Time: Thursday, September 6, 16:30 - 18:30
Presentation Time:Thursday, September 6, 16:30 - 18:30
Presentation: Poster
Paper Title: Deep Graph Convolutional Neural Networks Identify Frontoparietal Control and Default Mode Network Contributions to Mental Imagery
Manuscript:  Click here to view manuscript
Authors: Michael Craig, Ram Adapa, Ioannis Pappas, David Menon, Emmanuel Stamatakis, University of Cambridge, United Kingdom
Abstract: Brain network connectivity has been characterized during a variety of tasks and neurological/psychiatric diseases. Recently, various machine learning methods, including artificial neural networks, have used connectivity measures to predict cognitive or disease state. One area where these methods could be useful is in the prognosis of patients with disorders of consciousness (DOC). Previous work has used mental imagery tasks to assess DOC patient volitional ability, however little work has focused on incorporating machine learning methods to automatically detect awareness in these patients. The present study aims to establish a baseline for these methods in classifying mental imagery states. We developed a graph convolutional network classifier that can distinguish between mental imagery states in healthy subjects using only functional connectivity data. Furthermore, we examined whether certain large scale brain networks were more predictive than others, and found that frontoparietal control and default mode networks were most predictive of whether a participant was performing a mental imagery task or resting. These results demonstrate that graph convolutional networks could be developed to aid in detection of awareness in DOC patients and show that changes in connectivity patterns in frontoparietal control and default mode networks underlie alterations in mental imagery.