Prediction of Post-stroke Motor Recovery Benefits from Measures of Sub-acute Widespread Network Damages
Elvira Pirondini1, Maria Giulia Preti2, Dimitri Van De Ville3, Adrian Guggisberg4, Cyprien Rivier5
1University of Pittsburgh, 2Center for Biomedical Imaging (CIBM, Switzerland), 3Medical Image Processing Laboratory, EPFL, 4University Hospital of Berne, Inselspital, 5Yale University
Objective:
Developing a more accurate post-stroke motor recovery prediction model leveraging brain network connectivity measurements.
Background:
Following an acute ischemic stroke, some patients recover almost completely, while others barely recover. It is known that lesion volume, initial motor impairment, and cortico-spinal tract asymmetry (benchmark features) significantly impact motor changes over time. However, accurate prediction models of patients’ recovery are still missing. Recent work suggested that disabilities arise not only from focal structural changes but also from widespread alterations in inter-regional connectivity. Here, we built on this hypothesis to develop a more accurate post-stroke motor recovery prediction model.
Design/Methods:
We evaluated a dataset of 37 patients (17 women) with upper extremity motor strokes. Their impairment was assessed at 2 weeks (baseline) and 3 months after the stroke using the Fugl-Meyer-Assessment. 18 graph measures of global structural connectivity were obtained using the patients’ T2 lesion mask to virtually reproduce the lesions in 60 healthy streamline tractography connectomes. We added these indirect measures to the benchmark features, and we used a ridge regression regularization in a leave-one-out cross-validation framework to predict motor recovery at 3 months post-injury.
Results:
Prediction accuracy significantly increased when using brain connectivity measures (R2=0.68, 95% CI: 0.64-0.74) as compared to benchmark features alone (R2=0.38, 95% CI: 0.31-0.51).
Conclusions:
We report that global measures of brain connectivity estimated by introducing virtual lesions in healthy structural connectomes significantly increase the accuracy of motor recovery prediction in stroke patients compared to focal and clinical measures used until now. This indirect estimation of the stroke impact on the whole brain connectome is more readily available than direct measures of structural connectivity obtained with magnetic resonance imaging (MRI) and would, therefore, have a more substantial clinical impact.