Page 82 - MEGIN Book Of Abstracts - 2023

impaired ability to respond to change is a core feature learning of support vector machines to classify 163
across dementias, and other conditions driven by brain MCI cases versus 144 healthy elderly controls from the
network dysfunction, such as schizophrenia. It validates BioFIND dataset. When using the covariance of planar
theoretical frameworks in which neurodegenerating gradiometer data in the low Gamma range (30-48 Hz),
networks upregulate connectivity as partially effec- we found that adding a MEG kernel improved classi-
tive compensation. The significance extends beyond fication accuracy above kernels that captured several
network science and dementia, in its construct valida- potential confounds (e.g., age, education, time-of-day,
tion of dynamic causal modeling (DCM), and human head motion). However, accuracy using MEG alone
confirmation of frequency-resolved analyses of animal (68%) was worse than MRI alone (71%). When simply
neurodegeneration models. concatenating (normalized) features from MEG and MRI
into one kernel (Early combination), there was no ad-
Keywords: Alzheimer's disease, bvFTD, dementia, dynamic vantage of combining MEG with MRI versus MRI alone.
causal modeling, mismatch negativity, multiple demand When combining kernels of modality-specific features
(Intermediate combination), there was an improve-
The Journal of neuroscience: the official journal of the ment in multimodal classification to 74%. The biggest
Society for Neuroscience (2022), Vol. 42, No. 15 (35260433) multimodal improvement however occurred when we
(1 citation) combined kernels from the predictions of modality-
specific classifiers (Late combination), which achieved
77% accuracy (a reliable improvement in terms of
Late combination shows that MEG adds to MRI in permutation testing). We also explored other MEG
classifying MCI versus controls (2022) features, such as the variance versus covariance of mag-
netometer versus planar gradiometer data within each
Vaghari, Delshad; Kabir, Ehsanollah; Henson, Richard N of 6 frequency bands (delta, theta, alpha, beta, low
gamma, or high gamma), and found that they generally
MRC Cognition and Brain Sciences Unit, University of Cam- provided complementary information for classification
bridge, UK; Department of Electrical and Computer Engineer- above MRI. We conclude that MEG can improve on the
ing, Tarbiat Modares University, Tehran, Iran; MRC Cognition MRI-based classification of MCI.
and Brain Sciences Unit, University of Cambridge, UK; Depart-
ment of Psychiatry, University of Cambridge, UK. Electronic Keywords: Alzheimer's disease, MEG, Machine learn-
address: [email protected] ing, Mild cognitive impairment, Multimodal integration,
Structural MRI
ABSTRACT Early detection of Alzheimer's disease (AD)
is essential for developing effective treatments. Neuro- NeuroImage (2022), Vol. 252 (35247546) (0 citations)
imaging techniques like Magnetic Resonance Imaging
(MRI) have the potential to detect brain changes before
symptoms emerge. Structural MRI can detect atrophy Sensitive and reproducible MEG resting-state
related to AD, but it is possible that functional changes metrics of functional connectivity in Alzheimer's
are observed even earlier. We therefore examined the disease (2022)
potential of Magnetoencephalography (MEG) to detect
differences in functional brain activity in people with Schoonhoven, Deborah N; Briels, Casper T; Hillebrand,
Mild Cognitive Impairment (MCI) - a state at risk of Arjan; Scheltens, Philip; Stam, Cornelis J; Gouw, Alida A
early AD. We introduce a framework for multimodal
combination to ask whether MEG data from a resting- Department of Clinical Neurophysiology and MEG Center,
state provides complementary information beyond Department of Neurology, Amsterdam Neuroscience, Vrije
structural MRI data in the classification of MCI ver- Universiteit Amsterdam, Amsterdam UMC, Amsterdam, The
sus controls. More specifically, we used multi-kernel Netherlands. [email protected]; Alzheimer

 ontents  Index 61
C