Individual identification for different age groups using functional connectivity strength

Background and purpose Many studies demonstrate individual differences in functional network, especially those with age. Meanwhile, aging is one of the potential risk factors for Alzheimer’s disease. Therefore, it is important to explore the discrepant pattern in aging population. Methods Most existing methods mostly target ancient atlas for the extraction of the classification features and not consider the effect of global signal. We use two novel atlases for the extraction of classification features and then use the whole and intra-hemispheric functional connectivity strength (FCS) as classification parameters to classify different age groups, respectively. Meanwhile, the regression of global signal or not during the preprocessing has been considered. Next, the support vector machine-recursive feature elimination (SVM-RFE) method is applied for feature selection and the SVM method is applied for classification. In addition, the receiver operating characteristic curve and area under the curve are drawn to evaluate the robustness of classifier. Finally, the discriminative features are related to the physiological mechanism of aging. Results The promising classification performance exhibits that the FCS can effectively distinguish different age groups. Moreover, the SVM-RFE method can increase the accuracy and extract the discriminative features. The classifiers constructed by the features derived from different atlas receive similar classification performance. Conclusion This study successfully distinguishes the young group, middle-aged group, and elderly group through FCS parameter, indicating the functional pattern of the network exists difference between three groups. Moreover, the results received by the SVM-RFE method and SVM classifier have the very good robustness and not specific to particular atlas and not affected by global signal and appropriate for the FCS of the whole brain or intra-hemisphere, which suggests that we can apply them to disease diagnosis in the future.