| ObjectivesAcupuncture is a safe and effective measure for the management of functional dyspepsia(FD).However,the efficacy of acupuncture shows significant difference in FD patients for their different physical states.The combination of machine learning techniques and multidimensional brain-gut features provides a promising approach to characterize the physical states of FD patients at the individual level and to predict the efficacy of acupuncture treatment accurately.In this study,we aimed to(1)classify FD patients from healthy subjects(HS)and functional constipation(FC)based on machine learning algorithm and functional brain network(FBN)features,and then select the important identifying features of FD patients;(2)predict the response to acupuncture as well as the improvement of symptoms of FD patients after treatment based on machine learning algorithm and FBN features/brain-gut multidimensional features,and then select the critical predictive features;(3)investigate the mechanisms of acupuncture for the effective treatment of FD by comparing the changes in identifying and predictive features before and after treatment between the high responders and low responders.MethodsOne hundred and twenty-seven FD patients,91 FC patients,and 92 HS were included in this study.(1)In the first part of this study,89 FD patients,91 FC patients,and 92 HS were included.We first extracted 35 independent components(IC)based on independent component analysis and constructed the functional connectivity matrices between each pair of ICs for all subjects,generating the FBN features.Then a classification analysis between FD patients and HS was conducted based on the FBN features,the‘increasing edge’feature selection method and support vector classification(SVC)algorithm.After this,the discrimination features of FD patients were extracted as region-of-interests(ROI),and the classification analysis between FD patients and FC patients was performed based on those ROIs and SVC algorithm to test the specificity of these FBN features to discriminate FD patients.Finally,the correlation analysis between the discrimination features and clinical symptoms was conducted.(2)In the second part of this study,we first conducted the prediction analysis for 89 FD patients based on the disease-specific FBN features and SVC algorithm,to predict the response degree of FD patients receiving a 4-week acupuncture treatment period.And then,we performed the prediction analysis based on the SVC algorithm,the SVR algorithm,and the interested FBN features which generated with‘filtering’feature selection method,to predict the response degree as well as symptom improvement of Symptom Index of Dyspepsia(SID)score and Nepean Dyspepsia Symptom Index(NDSI)score.The selected FBN features were regarded as the predictive features.Thirdly,we applied the brain-gut multidimensional features which contained FBN features,functional brain activity features,demographic features,clinical symptom features,and psychological features to establish the prediction models,so as to predict the response degree as well as improvement of SID score and NDSI score.Fourthly,we included 38samples as the independent validation set to validate the generalization ability of the prediction models.(3)In the third part of this study,65 FD patients were included to explore the similarities and differences of the effects of acupuncture treatment on the disease-identification and efficacy-prediction FBN features.There were four steps of comparison,the comparison between pre-and post-treatment FBN features in all FD patients,in high responders,and in low responders,as well as the inter-group comparison of pre-and post-treatment FBN features changes between high and low responders.Moreover,correlation analysis between the changed FBN features and clinical symptom improvement was conducted.Results(1)The results of classification analysis showed that the 20 edges included caudate-superior orbitofrontal gyrus(IC33-IC31),thalamus-parahippocampal gyrus(IC69-IC37),and superior orbitofrontal gyrus-superior orbitofrontal gyrus(IC15-IC31)were identified as the discriminating features,which could effectively distinguish FD patients from HS.The average accuracy rate of classification in the training-test set was84.43%,and in the independent verification set was 80.14%.Moreover,these edges were also found to be effective to distinguish FD patients from FC patients with an accuracy rate of 91.7%.The results of correlation analysis suggested that the precentral gyrus related edges were significantly correlated with the SID score in FD patients.(2)Results of prediction analysis for acupuncture efficacy showed that 1)the response degree of FD patients could not be predicted with the disease-specific FBN features.2)the response degree as well as the improvement of SID score could be predicted with the precuneus-anterior cingulate cortex(IC55-IC79),caudate-precentral gyrus(IC33-IC6),superior marginal gyrus-anterior insula(IC49-IC28),and other 35 FBN features.Namely,the high responders and low responders could be identified with an average accuracy rate of76.32%,and the improvement of SID score could also be predicted with an R~2of 0.24.3)Using the brain-gut multidimensional data as predictive features,the performance of prediction models would significantly improve.The average accuracy rate of classification between high responders and low responders was increased to 88.04%,and the prediction for the improvement of SID score obtained an average R~2of 0.44.4)The prediction model had good generalization.In the independent verification set,the classification accuracy rate between high responders and low responders was 71.1%,and the predication of SID score improvement achieved an R~2of 0.39.(3)Seventeen discriminating features and 22predictive features were changed after 4-week acupuncture treatment.The discriminating features,the predictive features,and the target features of acupuncture treatment for FD patients were overlapped in the orbitofrontal grays.The changes of these edges related to the thalamus and caudate were more significant in the high responders compared to the low responders.Moreover,the changes of functional connectivity of thalamus-supplementary motor area,caudate-superior frontal gyrus and caudate-middle frontal gyrus were significantly positively correlated with the improvement of patients’clinical symptoms.Conclusion(1)The functional connectivity between the thalamus,caudate and the orbitofrontal cortex,sensorimotor cortex,parahippocampus were the effective FBN features to identify FD patients from HS and FC patients.The regulation to the functional connectivity of the thalamus and caudate was the potential mechanism for acupuncture treating FD with higher efficacy.(2)The functional connectivity of orbitofrontal gyrus,the core node of the reward loop,was the key feature that could effectively predict patients’response to acupuncture as well as the improvement of clinical symptoms after treatment.Moreover,the orbitofrontal gyrus was the overlapped feature of FD identification,acupuncture effects prediction,and realization of acupuncture effects.(3)Based on the"demographic characteristics-clinical symptom score-emotional mental state-brain spontaneous activity-functional brain network"features,the performance of the prediction models could significantly improve. |
PDF Full Text Request