Audio-visual crossmodal fMRI connectivity differentiates single patients with disorders of consciousness

• ¹ University of Liège, Cyclotron Research Center & CHU Neurology Department, Belgium
• ² Weill Cornell Medical College, Department of Radiology and Citigroup Biomedical Imaging Center, United States
• ³ University of Salzburg, Department of Psychology and Centre for Neurocognitive Research, Austria
• ⁴ Christian-Doppler-Clinic, Paracelsus Private Medical University, Neuroscience Institute and Centre for Neurocognitive Research, Austria
• ⁵ Christian-Doppler-Klinik, Paracelsus Private Medical University, Department of Neurology, Austria
• ⁶ Weill Cornell Graduate School of Medical Sciences, Department of Neuroscience, United States
• ⁷ Weill Cornell Medical College, Department of Neurology and Neuroscience, United States
• ⁸ MIT, Martinos Imaging Center at McGovern Institute for Brain Research, United States
• ⁹ Universidad Central de Colombia, Computer Science Department, Colombia
• ¹⁰ University of Liège, Cyclotron Research Center, Belgium
• ¹¹ CHU University Hospital, Department of Radiology, Belgium

INTRODUCTION Despite advances in resting state fMRI (rsfMRI) investigations in human consciousness, clinicians remain with the challenge of how to implement the rsfMRI paradigm on an individualized basis in the clinical setting (1). We here assessed the clinical relevance of the rsfMRI paradigm in patients with disorders of consciousness with the aim to: * describe group-level functional connectivity in large-scale networks * determine the anatomical correlate of each network in terms of the behavioral characteristics measured with the Coma Recovery Scale-Revised (CRS-R2) * identify the network which best differentiates between the groups of non-communicating patients in minimally conscious state (MCS) and vegetative state/unresponsive wakefulness syndrome (VS/UWS) * based on this network’s functional connectivity, to automatically differentiate patients in MCS and VS/UWS at the single-subject level METHODS - Participants: Patients in MCS, VS/UWS, coma (15 women; mean age 49±18 yrs, 16 traumatic, 31 non-traumatic of which 13 anoxic; 35 patients assessed in the chronic setting, i.e. >1 month post-insult) and healthy volunteers (8 women; mean age 45±17 yrs) scanned on 3T MRI machines in a sedation-free eyes-closed condition in 3 centers. -Data preprocessing (SPM8): slice time correction, realignment, segmentation of structural data, spatial and functional normalization into standard stereotactic MNI space and spatial smoothing (Gaussian kernel of 6mm FWHM). Further motion artifact detection and rejection was done using Artifact Detection tools (ART; http://www.nitrc.org/projects/artifact_detect). For noise reduction, we used the anatomical CompCor method (3) as implemented in CONN (http://www.nitrc.org/projects/conn), which models the influence of noise as a voxel-specific linear combination of multiple empirically estimated noise. Signals from the white matter (WM) and cerebrospinal fluid (CSF) noise regions of interest (ROIs) were extracted from the unsmoothed functional volumes to avoid additional risk of contaminating WM and CSF signals with gray matter signals. A temporal band-pass filter of 0.007–0.09 Hz was applied. Residual head motion parameters were regressed out. -Data analysis: Group-level seed-based functional connectivity was investigated for the default mode network (DMN; 10 ROIs), frontoparietal (15 ROIs), salience (23 ROIs), auditory (7 ROIs), sensorimotor (3 ROIs) and visual (6 ROIs) networks. Between-group inferential statistics and machine learning algorithms were used to identify the network with the best discriminative capacity between MCS and VS/UWS. Based on the functional connectivity of the identified network, a linear kernel Support Vector Machine classifier (regularization parameter C=1) was used to further differentiate patients on an individual basis. RESULTS Group-level seed-based functional connectivity of six large-scale networks was fully identified in healthy volunteers (n=21), well preserved in MCS (n=26), hardly identified in VS/UWS (n=19) and absent in coma patients (n=6). For each network, resting state functional connectivity positively correlated with the level of consciousness as measured by the CRS-R. The feature extraction method (t-test) indicated the auditory network with the highest discriminative capacity between the groups. Independently assessed patients (test dataset n=22) were classified accurately (95%) based on the crossmodal auditory-visual interaction. DISCUSSION The so far resting state fMRI-based differentiation of patients in separate clinical categories has been performed either at the group-level (4,5) or concerned the classification between healthy and pathological groups (6). We here showed that intrinsic fMRI audio-visual connectivity can differentiate MCS from unresponsive patients with acceptable clinical accuracy. Our findings point to the significance of preserved top-down processes in minimal consciousness, seemingly supported by crossmodal audio-visual interaction and highlight the clinical utility of the resting paradigm for single-patient diagnostics.