EEG Source Spatial Neural Representations Of Trait Depression And Impulsiveness Predict Suicidal Ideation

Objective:Suicide is a global public mental health problem that endangers human life and health.Suicide is preventable,and the detection and prediction of suicidal ideation is very important for the early precaution of suicidal behavior.Beck’s cognitive model for suicide focuses on the cognitive processes related to suicidal behavior,and has theoretical guiding significance for the prediction of suicide ideation.Clinicians and public health staff mainly use traditional methods such as questionnaires or“question-answer-mode”clinical consultations for suicidal ideation prediction.It is effective for people who have obvious suicidal ideation.But for the health ones who have not yet expressed distinct thoughts but are at very high risk,the questionnaires and“question-answer-mode”methods may be disadvantageous because of strong subjectivity and unstable results.Exploring objective behavioral and cognitive neural representations is an important approach to improve the accuracy and stability of suicidal ideation prediction.In addition,the traditional method of using statistical modeling to predict suicidal ideation cannot be well validated,while machine learning can enable the prediction results to be validated at a lower time cost.Based on Beck’s cognitive model for suicide,this study applies the method of psychological assessment to investigate the effects of two important factors,namely,impulsiveness and trait depression,on suicidal ideation.Based on these,we concretize the cognitive model for suicide and explore the cognitive neural representations of impulsiveness and trait depression in different dimensions through behavioral experiments,EEG sourcing and functional network analysis.Then we use these representations as predictive features to train a computational model for suicidal ideation to examine the predictive effect of the features on it.The results can validate Beck’s cognitive model for suicide at the level of EEG sourcing space,which can provide new evidence for early precaution of suicide behavior.Subjects:（1）Path model analyses:Network analysis—274 valid subjects;Moderated mediation path model analyses—480 valid subjects.（2）ERP and EEG source spatial neural representations analyses:116 valid subjects.（3）Machine learning:Training dataset—928 samples;Internal testing dataset—232samples;External testing dataset—217 samples.Methods:（1）To investigate the relationships among different dimensions of trait depression and impulsiveness,network analysis method was applied.The effects of different dimensions of trait depression and impulsiveness on suicidal ideation were investigated through path analyses.（2）The Iowa gambling task paradigm was applied to investigate the ERP,EEG sourcing and source spatial functional network neural representation variances for reward and punishment feedbacks of different dimensions of high and low trait depression and impulsiveness groups.Repeated measurement of variance analysis and generalized estimating equations were used to compare the group variances of the ERP and EEG source indexes,respectively.Independent-sample t test was conducted to compare the group variances of the EEG source spatial functional network indexes.（3）The EEG source spatial neural representations were used as features.The jointly classification of trait depression and impulsiveness was used as a response.We trained four machine learning models,namely,optimized decision tree,optimized naive Bayes,optimized support vector machine and RUSBoost.We then compared the models’performance,selected the optimal model,and applied it to the prediction of suicidal ideation to test the accuracy.Results:（1）Anhedonia was associated with nonplanning（W=0.193）and cognitive impulsiveness（W=0.157）.Dysthymia was associated with nonplanning（W=0.089）and motor impulsiveness（W=0.309）.Motor impulsiveness affected suicidal ideation,mediated by dysthymia(β_mediation=0.311,P<0.001).Nonplanning impulsiveness and cognitive impulsiveness were mediated by anhedonia in affecting suicidal ideation(β_mediation=0.334/0.365,P<0.001/0.001).Nonplanning and cognitive impulsiveness could reinforce anhedonia’s mediation effect(β_nonplanning=0.109,P=0.013;β_cognitive=0.126,P=0.010).（2）ERP（1）Positive feedbackThe N100(F_{nonplannig OZ}=3.328,P=0.071 marginal),P200(F_{nonplannig OZ}=6.032,P=0.016)and N200(Fdysthymia FCZ=4.310,P=0.040;Fdysthymia CZ/PZ=3.829/3.176,P=0.053/0.077 marginal;F_{nonplannig OZ}=2.991,P=0.086 marginal)components had greater amplitudes among high scored groups.The differences in the N100 a nd P200 components were mainly located on the posterior scalp.The difference in the N200 component were located on the middle and anterior scalp.The P300（Fnonplannig OZ=4.624,P=0.034;Fdysthymia PZ=3.946,P=0.049;Fdysthymia FCZ/CZ=3.466/3.740,P=0.066/0.056 marginal）and LPP(F_{nonplannig OZ}=3.287,P=0.072marginal;F_{dysthymia FCZ/CZ/PZ}=4.714/4.373/5.286,P=0.032/0.039/0.023;F_{dysthymia FZ/CPZ/POZ}=2.773/3.644/2.816,P=0.099/0.059/0.096 marginal)components in low impulsive and low trait depressive people were larger in amplitudes and were mainly located on the middle and anterior scalp.（2）Negative feedbackThe N100(F_{nonplanning FPZ}=5.596,P=0.020;F_{nonplanning CZ}=3.335,P=0.070 marginal;Fdysthymia POZ=5.158,P=0.025;Fanhedonia OZ=3.067,P=0.083 marginal)and P200(Fmotor _OZ=2.881,P=0.092 marginal)components in low impulsive and low trait depressive people had greater amplitudes and were mainly located on the posterior scalp.The N200(F_{cognitive FZ/FCZ/CPZ/POZ}=3.200/2.939/3.715/3.403,P=0.076/0.089/0.056/0.068 marginal;F_{anhedonia FCZ}=3.290,P=0.072 marginal)component in low impulsive and low trait depressive people also had greater amplitudes and were mainly located on the middle and frontal scalp.The P300（Fcognitive FZ/POZ=4.189/6.795,P=0.043/0.010;Fcognitive FCZ/CZ/CPZ/PZ=3.487/3.238/3.911/2.862,P=0.064/0.075/0.050/0.093 marginal）and LPP(F_{cognitive FCZ}=3.264,P=0.073 marginal)components in high impulsive and high trait depressive people had larger amplitudes and were mainly located on the anterior scalp.（3）EEG sourcing（1）Positive feedbackHigh impulsive and high trait depressive people had weaker early visual processing.The main differential processing brain areas were left superio r frontal sulcus（Wχ2anhedonia N100=4.122,P=0.042）,right superior occipital gyrus(Wχ² anhedonia N100=4.102,P=0.043;Wχ²_{cognitive N100}=4.732,P=0.030)and right precentral gyrus(Wχ²_{nonplaning P200}=4.345,P=0.037).High anhedonic individuals had larger advanced cognitive processing.The main brain area were dorsal post cingulate gyrus(Wχ²_{anhedonia N200}=4.297,P=0.038),right operculum of inferior frontal gyrus(Wχ²_{anhedonia N200}=7.158,P=0.007),right triangulum of the inferior frontal gyrus(Wχ²_{anhedonia P300}=6.476,P=0.011;Wχ²_{anhedonia LPP}=8.558,P=0.003).Low nonplanning impulsive individuals had larger advanced cognitive processing of positive feedback information.The main brain area were right middle frontal gyrus(Wχ²_{nonplanning P300}=4.234,P=0.040;Wχ²_{nonplanning LPP}=5.567,P=0.018)and right operculum of inferior frontal gyrus(Wχ²_{nonplanning LPP}=4.058,P=0.044).（2）Negative feedbackEarly visual processing was weaker in high impulsive and high tra it depressive people,with the main differential brain area of inferior part of right precentral sulcus(Wχ²_{dysthymia N100}=11.917,P=0.001),superior part of right precentral sulcus(Wχ²_{dysthymia N100}=8.130,P=0.004),and left superior occipital gyrus(Wχ²_{motor P200}=4.873,P=0.027).High impulsive and high trait depressive people had stronger advanced cognitive processing.The main brain areas were located in cingulate cortex,including left ventral posterior cingulate gyrus(Wχ²_{dysthymia N200}=4.896,P=0.027;Wχ²_{anhedonia N200}=6.054,P=0.014;Wχ² _{nonplanning N200}=5.891,P=0.015),left dorsal posterior cingulate(Wχ²_{anhedonia N200}=6.308,P=0.012),right anterior cingulate gyrus and sulcus(Wχ_nonplanning²P300=13.640,P<0.001;Wχ²cognitive P300=8.164,P=0.004;Wχ² nonplanning LPP=7.313,P=0.007;Wχ² _{cognitive LPP}=6.895,P=0.009).（4）EEG source spatial functional network（1）Positive feedbackFor the ECN network,the indexes of the node clustering coefficient（NCC）and node local efficiency（NLE）on left inferior frontal gyrus during 250ms and 350ms of the high anhedonia group were higher than the low anhedonia group(t _NCC=3.476,P=0.0007;t _NLE=3.365,P=0.001).The indexes of the NCC and NLE on left anterior middle cingulate gyrus during 300ms and 450ms of the high anhedonia group were higher than the low anhedonia group(t _NCC=3.232,P=0.002;t _NLE=3.406,P=0.0009).The indexes of the NCC and NLE on left operculum of inferior frontal gyrus during 600ms and 800ms of the high cognitive impulsiveness group were higher than the low cognitive impulsiveness group(t _NCC=3.356,P=0.001;t _NLE=3.481,P=0.0007).For the FPN network,the indexes of the NCC on left operculum of inferior frontal gyrus during 250ms and 350ms of the high anhedonia group were higher than the l ow anhedonia group（t=3.172,P=0.002）.The indexes of the NLE on left operculum of inferior frontal gyrus during 300ms and 450ms of the high nonplanning impulsiveness group were higher than the low nonplanning impulsiveness group（t=3.152,P=0.002）.（2）Negative feedbackFor the FPN network,the indexes of the node degree centrality（DC）on right inferior precentral sulcus during 250ms and 350ms of the high anhedonia group were lower than the low anhedonia group（t=-3.924,P=0.0004）.The indexes of the DC on left angular gyrus during 250ms and 350ms of the high cognitive impulsive group were lower than the low cognitive impulsive group（t=-3.795,P=0.0002）.The indexes of the DC on right operculum of inferior frontal gyrus and left angular gyrus during 300ms and 450ms of the high cognitive impulsiveness group were lower than the low cognitive impulsiveness group（t=-2.732/-3.008,P=0.007/0.003）.The indexes of the NCC and NLE on right operculum of inferior frontal gyrus during 300ms and 450ms of the high cognitive impulsiveness group were lower than the low cognitive impulsiveness group（t=-3.521/-3.596,P=0.0006/0.0005）.（5）The 7 key EEG source spatial neural representations related to dysthymia and nonplanning impulsiveness could predict suicidal ideation wit h an AUC of 0.72.The 8key EEG source spatial neural representations related to anhedonia and cognitive impulsiveness could predict suicidal ideation with an AUC of 0.75.The most important predictive features were the cognitive neural representations of negative feedbacks,which were mainly concentrated in the three time-windows:250～350ms,300～450ms and600～800ms.The mainly source spatial cortical activations presented in cingulate cortex,insular cortex,operculum of inferior frontal gyrus,transverse frontopolar gyrus and sulcus and the functional network features of operculum of inferior frontal gyrus.Conclusion:（1）Motor impulsiveness both directly and indirectly affects suicidal ideation through dysthymia.Nonplanning impulsiveness and cognitive i mpulsiveness affect suicidal ideation through the mediation of anhedonia.Simultaneously,both have positive moderation effects on the mediation effects.（2）High impulsiveness and high trait depression individuals have weaker early visual processing of positive feedbacks,but different variance modes on different dimensions in advanced cognitive processing.However,they have weaker early visual processing of negative feedbacks,while larger advanced cognitive processing.The main advanced cognitive processing brain areas for positive feedbacks are cingulate cortex,operculum and triangulum of inferior frontal gyrus,and middle frontal gyrus.The main advanced cognitive processing brain areas for negative feedbacks is cingulate cortex.The different effects of operculum of inferior frontal gyrus,cingulate gyrus,and inferior part of precentral sulcus in the ECN and FPN functional networks are the cognitive neural representations of trait depression and impulsiveness.（3）In response to negative information,cortical activations in cingulate cortex,insula cortex,operculum of inferior frontal gyrus,transverse frontopolar gyrus and sulcus and the functional network features of operculum of inferior frontal gyrus on specific time-windows of the individuals with trait depression and impulsiveness can accurately predict suicidal ideation.（4）When exposed to negative information and begin active cognitive processing,individuals with high anhedonia and cognitive impulsiveness have excessive cognitive control over threats in the early stage,and excessively weakened cognitive control in the middle stage,leading to excessive negative bias in the late stage.For people with high dysthymia and nonplanning impulsiveness,when they are exposed to negative information and begin active cognitive processing,due to the excessive negative processing bias in the late stage,they develop negative emotions,which could be the cognitive neural processing basis of suicidal ideation.