fMRI-based inverse analysis of stroke patients´ motor functions

The feasibility of automating the evaluation of stroke chronic patients´ motor functions has been explored while analyzing their corresponding fMRI studies with statistical parametric analysis, statistical inference analysis and a nonlinear multivoxel pattern-analysis classifier based on a feed-forward backward-propagation neural network. After doing principal component analysis and independent component analysis on an fMRI image data set, acquired after technology-based rehabilitation sessions of patients after stroke, an artificial neural network is trained with noncorrelated independent image-parameter vectors to discriminate statistical patterns of brain activations corresponding to each of the target sign language-like primitive hand movements that patients performed in the fMRI scanner while motor stimuli were being presented. The results look so promising that building a rehabilitation prognostics system could be looked forward.