Resting-state FMRI Data Analysis Based On The Hidden Markov Model

Functional magnetic resonance imaging(f MRI)is a technology that indirectly measures neural processing in the brain based on Blood Oxygen Level Dependent signal.It can be used to identify the spatial distribution network in the brain,providing efficient and fast advanced technology for brain function detection.The resting state is when the subject remains awake,relaxed,and avoids any specific mental activity.Functional imaging of the brain in the resting state is defined as resting-state f MRI,which reflects the spontaneous activity of the human brain.It has been shown that the spontaneous activity is highly organized.The Hidden Markov Model can be used to simulate brain activity,and the f MRI signals(observation sequence)are generated by a series of hidden states(sequence of states)that form a Markov chain.In recent years,resting-state f MRI data analysis based on the Hidden Markov Model is widely used in many fields,such as schizophrenia,autism,Alzheimer’s disease,and brain aging.However,most studies analyze the data of the multi-subject data of whole brain,and ignore the effects of subject specificity and the overlapping effects of different networks(the network specificity).In this paper,the subject specificity and network specificity is analyzed.This thesis first solves the learning problem by using the variational Bayesian algorithm,estimating the model parameters(the transition probability of the hidden state,initial probability,conditional probability).Then,this thesis uses the Viterbi algorithm to solve the Hidden Markov Model prediction problem decoding the hidden state sequence.Then drawing free energy change curves using the Kullback Leibler distance,to find the optimal number of states.Finally,this thesis builds mixed-effects models to calculate the intra-class correlation of the time parameters corresponding to the sequence of hidden states,exploring the test-retest reliability of resting-state f MRI data analysis based on the Hidden Markov Model.The research content of this thesis is mainly divided into two parts.One is the subject-specific study,which models the single-subject data and finds that the states of different subjects is different,so it is necessary to conduct a single-subject study.The test-retest reliability of the single subject was explored by calculating the intra-class correlation of time parameters,and the results showed that the test-retest reliability was good and the single subject study was feasible.The second is the network specificity study,considering the superposition effect of different network activities in the brain,and modeling the single network data.The results show that the single network test-retest reliability is good,and the single network study is feasible.