Page 278 - MEGIN Book Of Abstracts - 2023

Leodante; Jetly, Rakesh; Dunkley, Benjamin T indices of brain functioning to understand and differ-
entiate these debilitating conditions.
Neurosciences & Mental Health, SickKids Research Institute,
Toronto, ON, Canada. [email protected]; MacDonald Translational psychiatry (2021), Vol. 11, No. 1 (34088901) (2
Franklin OSI Research Centre, London, ON, Canada; Defence citations)
Research and Development Canada, Toronto, Canada;
Department of Surgery, University of Toronto, Toronto, ON,
Canada; Department of Psychiatry, Faculty of Medicine, Predicting PTSD severity using longitudinal
Dalhousie University, Halifax, NS, Canada; Department of magnetoencephalography with a multi-step
Medical Imaging, University of Toronto, Toronto, ON, Canada learning framework (2020)

ABSTRACT Post-traumatic stress disorder (PTSD) and Zhang, Jing; Wong, Simeon M; Richardson, J Don; Jetly,
mild traumatic brain injury (mTBI) are highly prevalent Rakesh; Dunkley, Benjamin T
and closely related disorders. Affected individuals often
exhibit substantially overlapping symptomatology - a Neurosciences and Mental Health, SickKids Research Institute,
major challenge for differential diagnosis in both mili- Toronto, ON, Canada; MacDonald Franklin OSI Research Cen-
tary and civilian contexts. According to our symptom tre, London, ON, Canada; Canadian Forces Health Services
assessment, the PTSD group exhibited comparable HQ, Ottawa, Canada; Department of Medical Imaging,
levels of concussion symptoms and severity to the mTBI University of Toronto, Toronto, ON, Canada
group. An objective and reliable system to uncover the
key neural signatures differentiating these disorders ABSTRACT Objective. The present study explores the
would be an important step towards translational effectiveness of incorporating temporal information in
and applied clinical use. Here we explore use of MEG predicting post-traumatic stress disorder (PTSD) sever-
(magnetoencephalography)-multivariate statistical ity using magnetoencephalography (MEG) imaging
learning analysis in identifying the neural features for data. The main objective was to assess the relationship
differential PTSD/mTBI characterisation. Resting state between longitudinal MEG functional connectome
MEG-derived regional neural activity and coherence (or data, measured across a variety of neural oscillatory
functional connectivity) across seven canonical neural frequencies and collected at two timepoints (Phase I
oscillation frequencies (delta to high gamma) were and II), against PTSD severity captured at the later time
used. The selected features were consistent and largely point.Approach. We used an in-house developed infor-
confirmatory with previously established neurophysi- matics solution, featuring a two-step process featuring
ological markers for the two disorders. For regional pre-learn feature selection (CV-SVR-rRF-FS, cross-valida-
power from theta, alpha and high gamma bands, the tion with support vector regression (SVR) and recursive
amygdala, hippocampus and temporal areas were iden- random forest feature selection) and deep learning
tified. In line with regional activity, additional connec- (long-short term memory recurrent neural network,
tions within the occipital, parietal and temporal regions LSTM-RNN) techniques.Main results. The pre-learn step
were selected across a number of frequency bands. This selected a small number of functional connections (or
study is the first to employ MEG-derived neural features edges) from Phase I MEG data associated with Phase II
to reliably and differentially stratify the two disorders PTSD severity, indexed using the PTSD CheckList (PCL)
in a multi-group context. The features from alpha and score. This strategy identified the functional edges
beta bands exhibited the best classification perfor- affected by traumatic exposure and indexed disease se-
mance, even in cases where distinction by concussion verity, either permanently or evolving dynamically over
symptom profiles alone were extremely difficult. We time, for optimal predictive performance. Using the
demonstrate the potential of using 'invisible' neural selected functional edges, LSTM modelling was used
to incorporate the Phase II MEG data into longitudinal
regression models. Single timepoint (Phase I and Phase

 ontents  Index 257
C