Research On Neural Mechanism Of Face Processing Using Neuroimaging

Face processing plays a very important role in human development and social communication.Given the importance and peculiarity of face processing,the investigation of the neural mechanism underlying face processing can facilitate to explore the basic mechanism of functional operation of brains.Thus,it is always the studying hot spot in the field of neuroscience and cognitive science.With the development of magnetic resonance imaging(MRI),this non-invasive neuroimaging methodology has been increasingly and generally used to explore the neural mechanism underlying face processing.Compared with the conventional methods,MRI can not only accurately and quantitatively measure and record the morphologies of brain tissues and neural activities,but also use the algorithms of image processing developed in the disciplines of computer science to analyze those brain features,thereby constructing the “brain-cognition”computing models and promoting the understanding of the functional segregation and integration of brain.The present study aimed to investigate the neural mechanism underlying face processing.To this end,the multi-modality MRI images were acquired in some elaborate cognitive experiments,based on which the machine learning models and multi-factor statistical models were constructed.Four key problems about the mechanism underlying face processing were explored in the present study.First,we decoded the category of the pure top-down processing of expertise(i.e.,face and Chinese character)elicited by a priori expectation based on the neural response patterns within the high-level visual cortex.The expectation can enhance the perception of the objects that are being expected.There are some category-selective regions within the high-level visual cortex.However,it is yet unclear whether the top-down processing of faces and Chinese characters can elicit category-selective neural response patterns,respectively.To address this problem,in the present study,the participants were required to maintain the face expectation and Chinese character expectation,with which they perceived faces and Chinese characters in the noise-only images,respectively.When the participants were performing the task,the functional MRI(f MRI)images were acquired.Then a 3D deep learning network model with 3D self-attention as the modulation and interpretation mechanism was constructed to discriminate the neural response patterns within the high-level visual cortex elicited by the pure top-down processing of faces from those of Chinese characters.The present study found that the neural response patterns elicited respectively by pure top-down face processing and Chinese character processing within the high-level visual cortex can be accurately classified.These findings suggested that the top-down processing elicited by a priori expectation can elicit the categoryselective neural response pattern.Such category-selective top-down modulation may be one of the neural mechanism underlying the influence of mind on perception.Second,we explored the dynamic integration of the top-down processing and bottom-up processing during the perception of faces and Chinese characters.According to the theory of predictive coding,the perceptual inference relies on both top-down and bottom-up processing.However,it is yet unclear how these two types of processing were integrated.To address this problem,in the present study,the participants were required to maintain the face expectation and Chinese character expectation,with which they perceived faces and Chinese characters not only in the noise-only images but also in the noise images containing faces and Chinese characters,respectively.When the participants were performing the task,the f MRI images were acquired.Then,an SVM classification model was constructed to discriminate the neural response patterns of whole brain regions between pure top-down face processing and Chinese character processing.This model was equal to a brain status that relied on the neural resources of the pure top-down processing to discriminate between faces and Chinese characters.Finally,the classification model was also tested using the neural response patterns elicited by the noise images containing faces and Chinese characters,respectively.This crossingclassification method can examine how the performance of the classification model varied when the actual information(i.e.,actual face or Chinese character)in the noise images increased,and thereby explore the influence of the predictability of the actual information input on the dynamic balance between top-down and bottom-up processing during the perception of faces and Chinese characters.The present study found that the performance of the classification model decreased with the increase of the actual information in the testing samples.This may be because that the top-down neural resources accordingly retreated when the perceptual inference was more and more dependent on the input of actual information.These finding revealed a dynamic trade-off balance between the top-down and bottom-up processing,which was modulated by the predictability of bottom-up input information.Third,we explored the neural mechanism underlying the inter-individual difference in face recognition ability.Although adults are all experts of face recognition,there is a significant amount of individual difference in their ability to recognize faces,which was found to be independent of the intelligence quotient of individuals.However,the exact neural mechanism underlying such inter-individual variation of face recognition ability is yet unclear.To address this problem,first,the face recognition ability was extracted using a regression method by filtering out the elements shared with the sub-level category and the basic category processing.Then an elastic net(E-net)model was constructed based on the white matter microstructural properties to predict the face recognition ability.The present study found that the face recognition ability can be accurately predicted by the Enet model based on the white matter integrity,particularly the six fiber tracts,namely the bilateral arcuate fasciculus(AF),the bilateral inferior longitudinal fasciculus(ILF),the callosum corceps minor(CFM),and the right cingulum cingulate.These findings not only confirmed the significant roles of the right ILF,but also revealed the bilateral AF and CMF mediating face recognition.Above all,because the CMF connects the homologous regions of bilateral hemispheres,our findings suggested the important role of hemisphere lateralization modulated by transcallosal connectivity in face recognition.Finally,we explored the neural mechanism underlying the other-race effect(ORE)of face processing.It has been generally accepted that own-race faces were recognized faster and more accurately than other-race faces.One of explanation of ORE is the more expertise of face processing with own-race faces than with other-race faces.However,the exact neural mechanism underlying ORE is yet unclear.To addressed this problem,in the f MRI experiment of the present study,the composite face effect and neural adaptation effect that respectively reflects the holistic face processing and the specialty of neural responses were combined.The multi-factor statistical models were constructed to explore the relationship between the race-related difference in face processing and the race-related difference in the activation of face-selective region within the high-level visual cortex.The present study found that the right fusiform face area(FFA)presented the sensitivity to the change in the face identity for both own-race faces and other-race faces.However,the right FFA only presented the composite effect for own-race faces,but not for otherrace faces,suggesting that along the pathway of the bottom-up face processing,own-race faces and other-race faces presented the holistic processing difference as early as when they were processed in the right FFA.This may be one of important reasons for ORE.