Poster No:
1445
Submission Type:
Abstract Submission
Authors:
Ardalan Aarabi1, Maedeh Khalilian1, Martine Roussel1, Olivier Godefroy1
Institutions:
1University of Picardy Jules Verne, Amiens, Picardie
First Author:
Co-Author(s):
Introduction:
Post-stroke deficits arise from both localized structural damage at the injury site and widespread network dysfunction caused by structural disconnection (Salvalaggio, 2020; Siegel, 2016). Recent research has leveraged alterations in functional and structural connectivity derived from lesions to predict behavioral deficits, accounting for 40–60% of variability in acute-stage stroke patients' functional scores [2]. In this study, we assessed the predictive power of lesion-derived structural network features, derived from structural disconnection, for predicting post-stroke motor, executive and processing speed deficits in stroke patients at the chronic stage.
Methods:
Structural imaging and clinical data from 340 stroke patients (aged 63.9±10.5 years) included in the GRECogVASC study cohort, who were experiencing motor, executive function, and/or processing speed deficits, were acquired six months post-stroke at Amiens University Hospital (Barbay, 2018). Lesions for each patient were manually segmented on 3D T1w images, normalized into the MNI152 template, and binarized to generate lesion masks. For post-stroke deficit prediction, structural features were extracted from structural disconnection maps. To this end, a probabilistic structural disconnection map was first generated for each patient using streamlines passing through the patient's lesion mask, estimated by fiber tracking from diffusion-weighted imaging data of over 400 healthy controls (62.87 ± 13.47 years) from the Cambridge Centre for Ageing and Neuroscience repository (CamCAN, Stage 2) (Taylor, 2017). In the probabilistic maps, each voxel represented a disconnection probability (0 to 1) based on the number of healthy subjects showing a disconnection in that voxel (Thiebaut de Schotten, 2011). Additionally, a connectivity matrix was initially constructed for each healthy subject based on the number of streamlines connecting parcels using a high-resolution parcellation atlas with 1133 regions. Then, for each patient, fibers passing through the patient's lesion were removed from the connectivity matrices of all healthy subjects and averaged to generate a group lesion-derived structural connectivity matrix for the patient. Finally, two graph metrics (degree and clustering coefficient) were calculated for each group connectivity matrix using graph analysis. Subsequently, each feature, including the lesion mask, probabilistic maps binarized at probability thresholds of 0.1, 0.3 and 0.5, lesion-derived group connectivity matrices, as well as nodal degree and clustering coefficient, underwent decomposition by PCA. The principal components explaining 99% of variance were then input into the ridge regression model using leave-one-out cross-validation across patients to predict post-stroke deficits in motor, executive function, and processing speed.
Results:
Our research indicated that lesion patterns, structural disconnection maps, and lesion-derived changes in structural connection strengths outperformed connectome-based features derived from brain networks. Optimal predictions for left/right motor scores (R2: 0.92 and 0.69) and processing speed scores (R2: 0.6) were achieved using the probabilistic structural disconnection map binarized at a probability threshold of 0.5. For executive function deficits, the most effective prediction was obtained by combining the probabilistic structural disconnection map binarized at probabilities of 0.1, 0.3, and 0.5, resulting in an R2 of 0.64.
Conclusions:
Our results indicate that although structural and functional connectivity can predict behaviors, they do not consistently surpass lesion-based models. Our study also revealed that models incorporating lesion-induced alterations in structural connection strengths slightly outperformed the lesion mask in predicting deficits. Structural disconnection patterns exhibited similar predictive capabilities to lesion masks, likely due to the alignment of structural disconnection patterns with the damage caused by the lesion.
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Connectivity (eg. functional, effective, structural) 2
Keywords:
Cerebrovascular Disease
Cognition
Computational Neuroscience
Data analysis
Modeling
Motor
STRUCTURAL MRI
Tractography
White Matter
Other - brain disconnectome
1|2Indicates the priority used for review
Provide references using author date format
Salvalaggio, A. (2020). Post-stroke deficit prediction from lesion and indirect structural and functional disconnection. Brain, 143(7), 2173-2188. doi:10.1093/brain/awaa156
Siegel, J. S. (2016). Disruptions of network connectivity predict impairment in multiple behavioral domains after stroke. Proc Natl Acad Sci U S A, 113(30), E4367-4376. doi:10.1073/pnas.1521083113
Barbay, M. (2018). Prevalence of Poststroke Neurocognitive Disorders Using National Institute of Neurological Disorders and Stroke-Canadian Stroke Network, VASCOG Criteria (Vascular Behavioral and Cognitive Disorders), and Optimized Criteria of Cognitive Deficit. Stroke, 49(5), 1141-1147.
Taylor, J. R. (2017). The Cambridge Centre for Ageing and Neuroscience (Cam-CAN) data repository: Structural and functional MRI, MEG, and cognitive data from a cross-sectional adult lifespan sample. Neuroimage, 144(Pt B), 262-269.
Thiebaut de Schotten, M. (2011). Atlasing location, asymmetry and inter-subject variability of white matter tracts in the human brain with MR diffusion tractography. Neuroimage, 54(1), 49-59.