Ridge penalized logistic partial least squares for predicting stroke deficit from infarct topography: A proof of concept study

To date, prediction of potential stroke deficit based on ischemic volumes has resulted in imprecise correlation with neurological outcome. This is due to the fact that information of infarct location is not incorporated into the model. We used a novel method, ridge penalised logistic partial least squares regression (RPL-PLS), to build a predictive model of neurological deficit from voxel involvement. The method identified the covariance between infarct locations and neurological deficits from stroke. It had high accuracy for predicting outcome in different neurological domains following stroke. This study provides proof of the concept that stroke outcome can be predicted from the information present in a MRI brain scan and paves the way for the development of similar models for understanding the neuroanatomy of neurological deficit and determining the outcome of rehabilitation and thrombolysis.