Research On The Dynamic Effective Connectivity And Brain Entropy For Bipolar Disorder Recognition And Risk Decision-making Behavior Prediction

The human brain is a finely divided and highly organized dynamic complex system.The dynamic Effective Connectivity(d EC)explored the dynamic interactions between different brain regions,while brain entropy represents the complexity of information processing in the brain itself over time.Bipolar Disorder(BD)is an affective psychiatric disorder with both manic and depressive episodes,characterized by a high misdiagnosis rate.BD patients show impaired decision-making function,impulsivity,self-injury,suicide,and other risky behaviors.No objective biomarkers were used as the standard for diagnosis and risk decision-making behavior evaluation of BD,which further increases the rate of misdiagnosis.Therefore,exploring its objective biomarkers plays an important clinical application value.Although current functional Magnetic Resonance Imaging(f MRI)studies have found that BD patients show abnormal d EC and entropy in brain regions,certain limitations are remained as following aspects.(1)The brain mechanism of risk decision-making behaviors in BD patients is unclear and there is a lack of effective prediction models for it.(2)The time scale of d EC is quite unitary,which would lead to a wealth of information about brain activity on different time scales being ignored.(3)Brain entropy is constructed as static brain entropy,and the relationships among the entropy of brain regions were ignored.Aiming to improve the accuracy of BD identification and reduce the prediction error of their risk decision behavior,based on the above aspects,this thesis constructed multi-time scale d EC,dynamic Brain Entropy,High-Order Brain Entropy(HOBEn),and entropy-based d EC(Entropy-d EC),providing scientific basis for clinical diagnosis.The main work of this thesis is as follows:1.Aiming to determine the brain pathological mechanism of risk decision-making behavior and construct an effective prediction model for it,based on the f MRI of Balloon Analog Risk Task(BART,Task-f MRI)and Resting-state f MRI(Resting-f MRI),the voxel-d EC(voxel-d EC)was constructed respectively.Two works were carried out:(a)The structural equation model was employed to construct the causal relationships among brain d EC,risk decision-making behavior,and clinical symptoms in BD patients;(b)Five regression algorithms were used to predict risk decision-making behaviors.Results indicated that d EC in the left middle temporal gyrus(resting state),and d EC in the left supplementary motor area(BART task)significantly influenced risk decisionmaking performances;the d EC in the left superior marginal gyrus significantly influenced that in the left superior frontal gyrus medial orbital margin,and that in the right cuneus,manic symptoms,and risk decision-making performances in BD patients.These results imply that risk decision-making behavior in BD patients may be driven by abnormal d EC in brain regions.Among five prediction models,Linear Regression obtained the best prediction performance,and this model could predict the impulsivity scale scores of BD,and could predict the risk decision-making and impulsivity scale score of Schizophrenia disorder(SZ),and could predict the impulsivity scale score of Attention Deficit Hyperactivity Disorder(ADHD).In conclusion,this work constructs the interaction relationship of brain-behaviors-symptoms and builts an effective prediction model for risk decision-making behavior in BD patients,providing auxiliary means for clinical intervention.2.To solve the defect in d EC time scale,this work coarse-grained the original time series of Task-f MRI and Resting-f MRI,and constructed new d EC(S1,S2,S3,S4)with scale factors of 1,2,3,4,as well as their concatenating feature,averaging feature1 and averaging feature2.The S1 is equivalent to the traditional d EC.Meanwhile,to comprehensively consider the information of the brain during task state and resting state,Relative Task-f MRI(RTask-f MRI)was constructed based on Task-f MRI and Restingf MRI.Four classification algorithms,including SVM,KNN,XGBoost,and LGBM,were used to identify BD patients and six regression algorithms were used to predict their risk decision-making behavior.The classification results showed that averaging feature1 of Task-f MRI and averaging feature2 of Resting-f MRI achieved the same best classification performances.Prediction results showed that S2 of RTask-f MRI achieved the best performance.The above results prove that the multi-time scale d ECs constructed in this work are more sensitive than the traditional d EC in identifying BD and predicting their behavior.In addition,the RTask-f MRI constructed in this work improves the predictive performance.The multi-time scale d EC provides a scientific reference for fully mining the hidden information of brain activity and is of great significance in quantifying the brain activity of patients with BD.3.Aiming to solve the lack of measurement for time variability of brain entropy,this work constructed four features based on Task-f MRI,namely traditional brain entropy,dynamic Brain Entropy,and their concatenating feature.Four classification algorithms were used to identify BD patients.Moreover,based on Task-f MRI,Restingf MRI,and RTask-f MRI,six regression algorithms were employed to predict the risk decision-making behavior of BD patients.The classification results showed that dynamic Brain Entropy achieved the best classification performance.The prediction results showed that the concatenating feature achieved the best results based on each of of three datasets.The above results indicate that the dynamic Brain Entropy improves the performance in the recognition and risk decision-making prediction of BD patients.4.Given the lack of methods to measure the interaction between different brain regions in complexity,this work constructed HOBEn and Entropy-d EC based on the traditional d EC and dynamic Brain Entropy.The data features include two parts:(a)cerebral cortex Folding index(Folding),traditional entropy,HOBEn(first-order,second-order,third-order,fourth-order),and their concatenating feature1,concatenating feature2;(b)Entropy-d EC(four orders as HOBEn)and their concatenating feature.In this work,SVM and KNN were used as classifiers to identify BD patients,and five regression algorithms were used to predict their risk decisionmaking behavior.Results showed that the second-order HOBEn achieved the best performance among all the HOBEn,and the third-order Entropy-d EC achieved the best performance among all Entropy-d ECs.In the prediction task,the concatenating feature of both HOBEn and Entropy-d EC achieved the best performances.The preformances of HOBEn and Entropy-d EC are better than Folding and traditional brain entropy in recognition and behavior prediction of BD patients,and the latter is better than the former.The High-Order Brain Entropy and Entropy-d EC constructed in this thesis provide important effective features for auxiliary diagnosis and behavior prediction in BD patients.