Multi-voxel patterns in the human brain associated with perceptual grouping of tactile frequencies

Introduction: As the use of mobile devices (particularly, wearable devices with vibrating alert features) are becoming more widespread, investigations on perceptual grouping of vibrotactile stimuli with different features, such as vibrating frequencies, are becoming more important for the design of effective haptic user interfaces. Previous psychophysical studies demonstrated that human perceive vibration frequencies as three distinctive groups: 'slow motion' ranging from 1 to 3 Hz, 'fluttering' ranging from 10 to 70 Hz, and 'smooth vibration' ranging from 100 to 300 Hz [1, 2]. This perceptual grouping pattern has been mainly explained based on the different characteristics of the tactile sensory innervations [3, 4]. However, characteristics of tactile innervations and sensory afferents do not fully describe perceptual grouping of vibrotactile stimuli. For instance, a boundary frequency should be between 40 and 50 Hz according to the afferent characteristics, but perception of vibrotactile stimuli is rather discriminated between 70 and 100 Hz. Furthermore, perceptual grouping is more likely to be affected by the neural encoding of vibration frequencies in the central nervous system, in addition to the characteristics of afferents. Here, we therefore search for the brain regions carrying frequency discriminative information using the searchlight multi-voxel pattern analysis (MVPA) [5] and compare the neural representations of different frequencies with the perceptual grouping patterns using multidimensional scaling (MDS). Methods: Fourteen subjects participated in this study and experimental procedures were approved by the Korea University (KU-IRB-11-46-A-1). Vibrotactile stimuli whose frequency varied from 20 to 200 Hz with an increment of 20 Hz were delivered to the tip of the index finger of the right hand by a vibrotactile stimulation device. Subjects performed ten runs of two sessions (one run for each frequency). Each session consisted of two consecutive periods: a 30 s resting period followed by a 30 s stimulation period. Functional images (T2*-weighted gradient EPI, TR = 3 s, voxel size = 2.0 × 2.0 × 2.0 mm3) were obtained using a 3T scanner. An information-based analysis with a cubical searchlight was employed to find spatially localized neuronal patterns varying with tactile frequencies. Decoding accuracies evaluated by a 2-fold cross-validation procedure were allocated to the center voxel of each searchlight. Then, we computed a correlation-based dissimilarity matrix and used MDS to map the neural representations for each of ten different frequencies onto the 2D space. Results: A random-effects group analysis revealed that a cluster exhibited statistically significant decoding capabilities to differentiate distinct frequencies (p50). This cluster covered the contralateral postcentral gyrus (S1) and the supramarginal gyrus (SMG). Mean decoding accuracy was 77.7 ± 13.8 and decoding accuracy results significantly exceeded the chance level (t13=7.5, p