Brain decoding of texture processing using independent component analysis and support vector machines

Abstract

Although brain activation in relation to tactile stimulation is well studied, the central processing of texture coding in humans is poorly understood. To explore such dynamics, we used a robotic setup to produce well-controlled stimuli consisting of two different textures (gratings of spatial period combinations 520+1920 mum and 400+1920 mum) which were moved across a subject's finger pad during electro-encephalograpy (EEG) acquisition. After decomposing the EEG signals with independent component analysis (ICA), a support vector machine (SVM) classifier was applied to each of the nonartifactual components to identify those where the temporal patterns encoded texture differences. In five of six subjects one such significant component was identified. Inverse source modeling revealed that in four of these subjects the components were located to the contralateral somatosensory cortex, consistent with previous research on tactile processing. In addition, components located to the ipsilateral somatosensory cortices not containing differentiating activity were identified. Thus, using state-of-the-art machine learning algorithms we demonstrated that it is possible to decode and localize subtle differences in temporal brain processing patterns related to textures perceived through the finger pad.