Study On Methods Of Brain State Modeling And Evolution Under Naturalistic Stimuli

Exploring the dynamic evolution of human brain states in natural scenes is not only a hot issue in brain science research,but is also of great significance to the research of brain inspired artificial intelligence algorithms.Previous studies on the dynamic changes of brain states were mainly based on discrete,repetitive and simple task stimuli or resting paradigms.However,these research paradigms lack ecological validity and cannot simulate continuous,dynamic and complex naturalistic stimuli in the real world.Therefore,it is also necessary to explore the effective methods of brain state modeling and evolution under naturalistic stimuli.In order to model the brain states under naturalistic stimuli,we used movie stimulated brain imaging data that close to the actual scene.In order to reduce the individual difference of brain activity between subjects,we proposed a dynamic hyperalignment(d HA)algorithm based on brain region level to functionally align the brain activity between subjects under naturalistic stimuli,and used hidden Markov model(HMM)to model the brain states.The results show that the proposed d HA-HMM algorithm framework significantly improves the consistency of functional activities among subjects,so as to mine the consistent HMM brain state among subjects.Based on the minimum free energy criterion,we found that there are 20 brain states switching dynamically during movie viewing,and different brain states reflect different combinations of large-scale brain networks.In order to reveal the mechanism of brain state evolution,we used the d HA-HMM framework to explore the correlation mechanism between brain state evolution and emotional changes induced by movie stimuli.Firstly,using multivariate analysis of variance,we confirmed that there was a multivariate relationship between HMM brain states and emotions,that is,HMM brain states were significantly grouped according to emotion labels in time domain.Then,based on the emotional sensitivity and specificity indicators,we quantitatively analyzed the multivariate relationship and found that the HMM brain states corresponding to the two opposite emotions of happiness and sadness also showed the opposite polarity.Furthermore,we analyzed the distribution of brain activation areas corresponding to brain states during happiness and sadness and found that strong activation was observed in the superior temporal gyrus,which confirmed that this area was related to the early process of emotional prosody processing.In addition to the traditional emotional processing areas,some advanced cognitive areas were also involved in emotional processing.When comparing the functional connections between brain states,it is found that there is a continuous strong functional connectivity among salience,cognitive,and sensorimotor networks during sadness,reflecting the interaction between high-level cognitive function network and low-level sensory network.Finally,using the data from different session of the same dataset,we verified the main results of the above research findings.In conclusion,this thesis uses the proposed d HA-HMM framework to dynamically model the brain state during naturalistic stimuli,quantitatively analyze its dynamic evolution process,and reveal that there are significant differences in brain activities and connection patterns between happiness and sadness.The proposed d HA-HMM framework can also be used in the study of other naturalistic stimuli and other acquisition modes.The findings promote our understanding of the evolution mechanism of brain states under naturalistic stimuli,expand our understanding of emotional cognitive mechanism in natural scene,and are also helpful to the field of emotional brain computer interface and affective computing.