Format

Send to

Choose Destination
See comment in PubMed Commons below
Cortex. 2014 Mar;52:35-46. doi: 10.1016/j.cortex.2013.11.005. Epub 2013 Nov 20.

Multiple fMRI system-level baseline connectivity is disrupted in patients with consciousness alterations.

Author information

1
Coma Science Group, Cyclotron Research Center & Neurology Department, University of Liège, Belgium. Electronic address: a.demertzi@ulg.ac.be.
2
Coma Science Group, Cyclotron Research Center & Neurology Department, University of Liège, Belgium; Computer Science Department, Universidad Central de Colombia, Bogotá, Colombia.
3
Neuroscience Institute and Centre for Neurocognitive Research & Department of Neurology, Christian-Doppler-Clinic, Paracelsus Private Medical University, Salzburg, Austria; Department of Psychology and Centre for Neurocognitive Research, University of Salzburg, Austria.
4
Coma Science Group, Cyclotron Research Center & Neurology Department, University of Liège, Belgium.
5
Department of Radiology, CHU University Hospital, University of Liège, Belgium.
6
Department of Anesthesiology, CHU University Hospital, University of Liège, Belgium.
7
Coma Science Group, Cyclotron Research Center & Neurology Department, University of Liège, Belgium; Brain & Mind Institute, Physics & Astronomy Department, Western University, London, Ontario, Canada.

Abstract

INTRODUCTION:

In healthy conditions, group-level fMRI resting state analyses identify ten resting state networks (RSNs) of cognitive relevance. Here, we aim to assess the ten-network model in severely brain-injured patients suffering from disorders of consciousness and to identify those networks which will be most relevant to discriminate between patients and healthy subjects.

METHODS:

300 fMRI volumes were obtained in 27 healthy controls and 53 patients in minimally conscious state (MCS), vegetative state/unresponsive wakefulness syndrome (VS/UWS) and coma. Independent component analysis (ICA) reduced data dimensionality. The ten networks were identified by means of a multiple template-matching procedure and were tested on neuronality properties (neuronal vs non-neuronal) in a data-driven way. Univariate analyses detected between-group differences in networks' neuronal properties and estimated voxel-wise functional connectivity in the networks, which were significantly less identifiable in patients. A nearest-neighbor "clinical" classifier was used to determine the networks with high between-group discriminative accuracy.

RESULTS:

Healthy controls were characterized by more neuronal components compared to patients in VS/UWS and in coma. Compared to healthy controls, fewer patients in MCS and VS/UWS showed components of neuronal origin for the left executive control network, default mode network (DMN), auditory, and right executive control network. The "clinical" classifier indicated the DMN and auditory network with the highest accuracy (85.3%) in discriminating patients from healthy subjects.

CONCLUSIONS:

FMRI multiple-network resting state connectivity is disrupted in severely brain-injured patients suffering from disorders of consciousness. When performing ICA, multiple-network testing and control for neuronal properties of the identified RSNs can advance fMRI system-level characterization. Automatic data-driven patient classification is the first step towards future single-subject objective diagnostics based on fMRI resting state acquisitions.

KEYWORDS:

Coma; Independent component analysis; Machine learning; Resting state; fMRI

PMID:
24480455
DOI:
10.1016/j.cortex.2013.11.005
[Indexed for MEDLINE]
PubMed Commons home

PubMed Commons

0 comments
How to join PubMed Commons

    Supplemental Content

    Full text links

    Icon for Elsevier Science
    Loading ...
    Support Center