Using FMRI To Study The Neural Representation Of Object Geometric Transformations In The Ventral Visual Pathway

When the shape of an object changes,the visual information is processed in the occipital,occipitotemporal,and temporal regions in the ventral visual pathway,and finally realizes the processing of the shape change and the recognition of the object identity.Previous studies proposed the representational transformation analysis method which uses linear transformation to describe multivariate pattern transformations in the ventral visual pathway during shape change,and they showed that there is no difference in the effect of linear matrix method for different degrees of changes(e.g.,rotation and scaling).In this thesis,we propose to further investigate the neural encoding principle of shape changes based on previous studies.In order to provide a formal description of shape changes,we use the Erlanger Programme as the mathematical framework for our study,based on the Theory of Topological Perception for shape perception.The main purpose of this study is to investigate whether the transformations in neural representations induced by different geometric transformations have a hierarchical relationship corresponding to the Erlanger Programme,and to investigate whether the linear method is adequate and necessary to explain the transformations in representations.In the thesis,by means of functional MRI and multivariate pattern analysis,we analyzed Euclidean and topological changes,which have different degrees of deformation.Firstly,within the whole-brain searchlight analysis,it was found that the linear matrix method,as well as the Procrustes analysis,which has a weaker degree of transformation than the linear matrix,provided a better prediction of Euclidean than topological changes in the visual region.Subsequently,three regions of interest(early visual cortex,ventral occipitotemporal,inferior temporal)were defined by the searchlight results and previous studies about transformation encoding in the ventral visual pathway,in which Procrustes analysis,linear matrix method,and neural network method that are more flexible than linear matrix were used.And it was found that,the more flexible the method used in the representation transformation analysis,the better the transformation effects were.And the prediction for Euclidean was better than topological in the Procrustes analysis and linear matrix method,the difference between Euclidean and topological was not significant in the neural network method.The results indicate that,the neural representation of different geometric transformations has a hierarchical structure similar to that in the Erlanger Programme,but does not correspond exactly to the changes in the visual stimulus.Linear matrix method is necessary but not sufficient,Procrustes analysis with weaker transformations cannot achieve effective analysis,and neural network with stronger transformations provides a better interpretation.It is suggested that both the degree of changes and the representation analysis methods have effects on the transformation effects,therefore,the research about neural representation of shape changes should take both the degree of experimental stimulus changes and the flexibility of the algorithms into account.