A Study On The Neural Mechanism Of Action Representation Based On FMRI

Human beings can perceive and understand the movement of others efficiently and accurately,so as to interact and communicate with the outside world.In contrast,the ability of the computers is far from the human brain,both in terms of efficiency and accuracy.Studying the neural mechanism underlying the action perception of human brain can not only provide a theoretical basis for computer simulation of human behavior perception,but also promote the development of computer vision and artificial intelligence.Functional magnetic resonance imaging(fMRI)has the advantages of noninvasive and high spatial resolution.It provides an effective tool for the study of cognitive neuroscience.With the help of f MRI technology,the study of neural mechanism underlying human action perception will greatly promote the study on computer simulation of feature extraction and the recognition of human behavior.In the current study,from the perspective of functional integration and functional localization,we used effective connectivity and multi-voxel pattern analysis to explore the neural interactions among three vital regions in occipitotemporal cortex during basic human action perception,and to identify the brain regions that can represent actions abstractly by generalizing across different visual formats and incidental visual features,so as to clarify the neural basis of action understanding.Firstly,using dynamic causal model analysis method,we selected three regions in occipitotemporal cortex as model nodes: EBA,hMT+ and pSTS.A model space with 576 models was constructed,and the models were estimated and selected.The fully connected models with reciprocal endogenous connectivity among the three regions were considered the most feasible.We found significantly enhanced modulated connectivity from h MT+ to both EBA and pSTS,as well as from EBA to pSTS.Our results revealed the manner in which the three regions interact during action recognition.We inferred that hMT+ would be the first region to show sensitivity to the presence of motion,and the activity of which would subsequently influence and be influenced by the other two regions.Then,human subjects were scanned while viewing four categories of human actions depicted in two distinct visual formats: the information-rich dynamic videos and the highly impoverished point light display.We thus used searchlight multi-voxel pattern analysis based on f MRI data to identify brain regions that can abstract action categories from different exemplars and the action formats.A validation analysis based on classification decoding was conducted subsequently to evaluate the reliability of the searchlight results.Our results revealed that actions were decodable across action formats in occipitotemporal cortex,inferior parietal cortex and some areas in frontal cortex.