Early prediction of long-term tactile object recognition performance after sensorimotor stroke.

ABSTRACT

Until now tactile agnosia has been reported only in small, but detailed cross-sectional case studies. Here we show that multi-voxel pattern analysis (MVPA) of early diffusion-weighted lesion maps can be used to accurately predict long-term recovery of tactile object recognition (TOR) in 35 subjects with varying hand skill impairment and associated specific daily activity limitation after cortical sensori-motor stroke. Multiple regression analysis revealed the essentially dysfunctional subprocesses for object recognition in the specifically impaired subjects, i.e., grasping as determined by a subtest of Jebsen Taylor hand function test, and perception of macrogeometrical object properties. The Gaussian process regression of MVPA represents a function that relates a selection of lesioned voxels as input variables to TOR performance scores as target variables. On the behavioural level, patients fell into three recovery subgroups, depending on TOR performance over the observation period. Only baseline motor hand skill and shape discrimination were significantly correlated with the TOR trajectories. To define functionally meaningful voxels, we combined information from MVPA of lesion maps and a priori knowledge of regions of interest derived from a data bank for shape recognition. A high significance for the predicted TOR performances over nine months could be verified by permutation tests, leading us to expect that the model generalises to larger patient cohorts with first cortical ischemic stroke. The lesion sites of the persistently impaired subjects exhibited an overlap with critical areas related to the MVPA prediction map in the cytoarchitectonic areas PFt of inferior parietal lobule and OP1 of parietal operculum which are associated with higher order sensory processing. This ultimate check corroborated the significance of the MVPA map for the prediction of tactile object recognition. The clinical implication of our study is that neuroimaging data acquired immediately after first stroke could facilitate individual forecasting of post-stroke recovery.