The Resting State Functional MRI Study Of Chemotherapy-related Subjective Cognitive Complaints In Breast Cancer Based On Machine Learning

Objective: To explore the value of screening resting state functional magnetic resonance imaging(rs-f MRI)parameters as features based on machine learning method in distinguishing breast cancer patients with postoperative chemotherapy-related subjective cognitive complaints(SCC)from postoperative non-chemotherapy(BC)and healthy controls(HC).Methods: This study recruited 40 subjects in the SCC group,49 in the BC group and 34 in the HC group,who underwent subjective report questionnaires(including Functional Assessment of Cancer Therapy-Cognitive Function and Beck depression inventory),objective cognitive assessment and rs-f MRI scanning.After data preprocessing,based the 90 regions of interest in anatomical automatic labeling brain atlas,the functional metrics of all three groups included functional connectivity(FC),amplitude of low frequency fluctuation(ALFF)and fractional ALFF(f ALFF),regional homogeneity(Re Ho),voxel-mirrored homotopic connectivity(VMHC)and degree centrality(DC)values were calculated and extracted as features set.Then,the features were selected by two sample t test(P < 0.01),removing variables with a high pairwise correlation(the correlation absolute threshold was set at 0.65)and least absolute shrinkage and selection operator regression to obtain the most discriminative features.At last,the support vector machine(SVM)models were built for classification(SCC vs.BC,SCC vs.HC)and permutation tests(5000 times)were performed.The receiver operating characteristic curve and area under curve(AUC)were used to evaluate the classification effectiveness of the model.Results: There were no significant differences between the SCC and BC groups in the Beck depression inventory and objective cognitive assessment scores.However,the SCC group had significantly worse subjective cognitive performance compared with the BC group.About the features selection and construct the SVM classification models:(1)Between the SCC and BC group,thirty-eight rs-f MRI features(mainly distributed by the brain regions of default mode network and subcortical network)included 25 FC,6 ALFF,2 f ALFF,4 Re Ho and one VMHC values were selected and the accuracy of the model was 82.0%(AUC=0.903,95%confidence interval: 0.822～0.956,P-value=0.0003);(2)Between the SCC and HC group,seventeen rs-f MRI features(mainly distributed by the brain regions of default mode network,frontal-parietal control network and cingular-opercular network)included 7 FC,2 ALFF,4 f ALFF,3 Re Ho and one DC values were selected and the accuracy of the model was 91.9%(AUC=0.943,95% confidence interval:0.863～0.983,P-value=0.0004).Conclusion: In this study,the SVM classifier constructed by using multi-level rs-f MRI imaging features screened by machine learning method could effectively distinguish breast cancer patients with SCC after postoperative chemotherapy from the BC and HC groups.And the selected rs-f MRI features provided objective neuroimaging evidence for understanding chemotherapy-related SCC,the early stage of chemotherapy-related cognitive impairment in breast cancer.