Page 280 - MEGIN Book Of Abstracts - 2023

and VA San Diego, UCSD Radiology Imaging Lab, 3510 (PTSD), an objectively measurable biomarker is highly
Dunhill Street, San Diego, CA 92121, USA; The Mind Research desirable; especially to clinicians and researchers.
Network, 1101 Yale Boulevard, Albuquerque, NM 87106, Macroscopic neural circuits measured using magneto-
USA; Department of Radiology, University of California, San encephalography (MEG) has previously been shown to
Diego and VA San Diego, UCSD Radiology Imaging Lab, 3510 be indicative of the PTSD phenotype and severity. In
Dunhill Street, San Diego, CA 92121, USA. Electronic address: the present study, we employed a machine learning-
RRLEE@UCSD.EDU based classification framework using MEG neural
synchrony to distinguish combat-related PTSD from
ABSTRACT Mild traumatic brain injury (mTBI) and post- trauma-exposed controls. Support vector machine
traumatic stress disorder (PTSD) are leading causes of (SVM) was used as the core classification algorithm.
sustained physical, cognitive, emotional, and behav- A recursive random forest feature selection step was
ioral deficits in the general population, active-duty directly incorporated in the nested SVM cross valida-
military personnel, and veterans. However, the under- tion process (CV-SVM-rRF-FS) for identifying the most
lying pathophysiology of mTBI/PTSD and the mecha- important features for PTSD classification. For the five
nisms that support functional recovery for some, but frequency bands tested, the CV-SVM-rRF-FS analysis se-
not all individuals is not fully understood. Conventional lected the minimum numbers of edges per frequency
MR imaging and computed tomography are generally that could serve as a PTSD signature and be used as the
negative in mTBI and PTSD, so there is interest in the basis for SVM modelling. Many of the selected edges
development of alternative evaluative strategies. Of have been reported previously to be core in PTSD
particular note are magnetoencephalography (MEG) pathophysiology, with frequency-specific patterns also
-based methods, with mounting evidence that MEG observed. Furthermore, the independent partial least
can provide sensitive biomarkers for abnormalities in squares discriminant analysis suggested low bias in the
mTBI and PTSD. machine learning process. The final SVM models built
with selected features showed excellent PTSD clas-
Keywords: Functional connectivity, GABA-ergic, Gamma sification performance (area-under-curve value up to
wave, Posttraumatic stress disorder, Slow wave, Traumatic 0.9). Testament to its robustness when distinguishing
brain injury individuals from a heavily traumatised control group,
these developments for a classification model for PTSD
Neuroimaging clinics of North America (2020), Vol. 30, No. also provide a comprehensive machine learning-based
2 (32336405) (4 citations) computational framework for classifying other mental
health challenges using MEG connectome profiles.

Classifying post-traumatic stress disorder using Scientific reports (2020), Vol. 10, No. 1 (32246035) (15
the magnetoencephalographic connectome and citations)
machine learning (2020)

Zhang, Jing; Richardson, J Don; Dunkley, Benjamin T Altered modulation of beta band oscillations during
memory encoding is predictive of lower subsequent
Neurosciences & Mental Health, SickKids Research Institute, recognition performance in post-traumatic stress
Toronto, ON, Canada. jing.zhang@sickkids.ca; MacDonald disorder (2020)
Franklin OSI Research Centre, London, ON, Canada; Depart-
ment of Medical Imaging, University of Toronto, Toronto, ON, Popescu, Mihai; Popescu, Elena-Anda; DeGraba,
Canada Thomas J; Hughes, John D

ABSTRACT Given the subjective nature of conventional National Intrepid Center of Excellence, Walter Reed Na-
diagnostic methods for post-traumatic stress disorder tional Military Medical Center, Bethesda, MD, United States;

 ontents  Index 259
C