Reply: Lesion network mapping predicts post-stroke behavioural deficits and improves localization

In their letter to the editor, Cohen and colleagues1 commented on our study2 in which we compared different structural and functional MRI methods to predict behavioural deficits in a large cohort of subacute stroke patients.3 Specifically, we estimated behavioural deficits in different domains (language, vision, motor, attention, and memory) using: (i) lesion location and size; (ii) structural disconnection of white matter pathways estimated in a healthy control atlas (SDC); (iii) functional disconnection of brain networks (FDC) estimated in healthy controls using an approach similar, but not identical to lesion-network mapping (LNM)4; and, finally, (iv) direct measures of functional MRI connectivity in a subset of patients. The prediction of behavioural impairments at 2 weeks post-stroke used ridge regression, a machine learning technique that provides out-of-sample estimates while minimizing the possibility of overfitting.5
Source: Brain - Category: Neurology Source Type: research