A Resting-state Functional Magnetic Resonance Study On Identification Of Patients With Major Depression Disorder And Bipolar Disorder Based On Machine Learning

Objective: Major depression disorder(MDD)and bipolar disorder(BD)are two affective disorders with a high rate of disability and fatality.It is difficult to distinguish the two disorders when BD patients experienced the first depressive episode.Followup study of MDD could effectively improve subjective diagnostic accuracy and restingstate functional MRI(rs-f MRI)could assist to identify the two diseases objectively.The functional differences of multiple brain regions were generally observed in the two diseases,and the study on the brain network may be helpful to find neuroimaging mechanisms of two diseases.In the properties of brain network,voxel-wise degree centrality(DC)is one of the most direct measures to describe the importance of functional network nodes and is sensitive to detect the functional differences between two diseases with a high test-retest reliability.Therefore,combining follow-up with DC will contribute to early identification of MDD and BD,however,the results of researches cannot be directly applied to clinical practice.A model building by machine learning can distinguish different individuals and be applied to a new data for classification or prediction.In this study,we intend to combine follow-up with DC to explore the specific and trait neuroimaging mechanisms of MDD and BD,which may provide neuroimaging basis for early identification of two diseases.Then an MDD-BD model with the best classification performance is established based on DC features,and be applied to a new data to test the ability of early identification of the patients who absence of confirmed symptoms.Methods: At baseline,MDD and BD patients were recruited and rs-f MRI was performed.Then MDD patients were followed up for more than 2 years.Those patients who developed mania or hypomania during follow-up were included in the BD before conversion,while those who never developed mania or hypomania were continued to be included in the MDD group.Finally,we recruited 128 BD patients(including 33 BD patients before conversion),110 MDD patients(including 33 patients who were age-,gender-,education-matched with BD patients before conversion)and 40 healthy controls(HCs)who were age-,gender-,education-matched with BD patients before conversion.Age,gender,education,course of disease,medication,total scores of 17-item Hamilton Rating Scale for Depression(HAMD-17),total scores of Hamilton Anxiety Scale(HAMA)and total scores of Young Mania Rating Scale(YMRS)of all subjects were collected.Demographic and clinical data were analyzed by using SPSS24.0 software.Based on MATLAB software platform,rs-f MRI data are preprocessed by DPARSFA(Data Processing Assistant for RBF Advanced)software package and analyzed by DPABI 4.2(Data Processing & Analysis for Brain Imaging Software Package Version 4.2).The study was divided into two parts.The voxel-wise degree centrality(DC)of BD patients before conversion,MDD patients and HC were compared in the first part.In the second part,firstly,the differences of DC between MDD and BD patients with definite diagnosis were compared;secondly,logistic regression classifier in the Easylearn software based on Scikit-Learn and classification features which were the brain regions with statistical DC differences were used to establish the MDD-BD classification model,and 5-fold cross-validation was used to test the model;finally,the model was applied to an independent test dataset(including BD patients before conversion and MDD patients).Results: 1.The first part included 33 MDD,33 BD patients before conversion and 40 HC,and there were no significant differences in age,gender and education among the three groups(P>0.05).The image results showed statistical DC differences in the middle occipital gyrus,superior temporal gyrus,middle temporal gyrus,inferior parietal gyrus,supramarginal gyrus,postcentral gyrus,superior frontal gyrus,middle frontal gyrus and inferior frontal gyrus(triangular part)of the right cerebrum among the three groups(P<0.05,GRF corrected).Post hoc analysis of DC values which were extracted from the above regions revealed that compared to HC and BD patients before conversion,MDD patients showed decreased DC values in the middle occipital gyrus,superior temporal gyrus and middle temporal gyrus of the right cerebrum(P<0.05),and there were no statistical differences between HC and BD patients before conversion(P>0.05).Compared with HC and MDD patients,BD patients before conversion showed increased DC values in the inferior parietal gyrus,supramarginal gyrus and postcentral gyrus of the right cerebrum(P<0.05),and there were no significant differences between HC and MDD patients(P>0.05).Compared with HC,patients with MDD and BD before conversion showed increased DC values in the superior frontal gyrus,middle frontal gyrus and inferior frontal gyrus(pars triangular)of the right cerebrum(P<0.05),and there were no statistical differences between patients with MDD and BD before conversion(P>0.05).2.The second part included 77 MDD patients and 95 BD patients,and there were no significant differences in age,gender and education between the two groups(P>0.05).In the comparison between MDD and BD patients,significant DC differences were observed in the bilateral superior frontal gyrus,bilateral medial superior frontal gyrus,bilateral orbital superior frontal gyrus,bilateral middle frontal gyrus,right orbital middle frontal gyrus,bilateral inferior frontal gyrus(opercular part),bilateral inferior frontal gyrus(triangular part),bilateral orbital inferior frontal gyrus,bilateral medial orbitofrontal gyrus,bilateral precentral gyrus,bilateral supplementary motor area,bilateral Rolandic operculum,bilateral paracentral lobule,bilateral superior parietal gyrus,bilateral inferior parietal gyrus,bilateral supramarginal gyrus,bilateral angular gyrus,bilateral precuneus,bilateral postcentral gyrus,bilateral superior occipital gyrus,right middle occipital gyrus,right inferior occipital gyrus,bilateral cuneus,bilateral lingual gyrus,bilateral calcarine gyrus,bilateral superior temporal gyrus,bilateral temporal pole: superior temporal gyrus,bilateral middle temporal gyrus,left inferior temporal gyrus,bilateral fusiform gyrus,bilateral insula,bilateral anterior cingulate gyrus,bilateral posterior cingulate gyrus,bilateral medial and paracingulate gyrus,right parahippocampal gyrus and right putamen(P<0.05,GRF corrected).Results of classification demonstrated that regions with statistical DC differences were defined as functional features,and a model with the best classification performance was established within these regions.The accuracy of the model is 0.80±0.04,the sensitivity is 0.80±0.04,the specificity is 0.78±0.11,AUC is 0.86±0.04.When the model was applied to the testing data,the accuracy is 0.55,the sensitivity is 0.61,the specificity is0.48,AUC is 0.54.Conclusion: 1.MDD patients showed specific decreased DC in the middle occipital gyrus,superior temporal gyrus and middle temporal gyrus of the right cerebrum as well as BD patients before conversion had specific increased DC in the inferior parietal gyrus,supramarginal gyrus and postcentral gyrus of the right cerebrum may provide specific and trait neuroimaging markers for the two diseases;the increased DC in the superior frontal gyrus,middle frontal gyrus and inferior frontal gyrus(pars triangular)of the right cerebrum may underlying the common neuroimaging mechanisms of the two diseases.2.Based on the DC features of the cortico-limbic-striatum neural circuit,the accuracy of the MDD-BD classification model was 80%,and the model identified 61%of BD patients before conversion who absence of confirmed symptoms.