Abstract
Judgment of face trustworthiness is an important social ability. In this study, we used linear discriminant analysis (LDA) and support vector machine (SVM) to classify the neural activation shown by older and younger adults to faces at different levels of trustworthiness. We first performed principle component analysis to extract features from fMRI data. Then we used a leave-one-subject-out approach in classifier training and testing. In the end, we used permutation testing to evaluate the significances of overall classification accuracy and each face category’s area under receiver operating characteristic curve (AUC), and used bootstrapped testing to compare classifier’s performance between OA and YA. Our results showed LDA and linear SVM could both differentiate the neural activation shown by both older adults (OA) and younger adults (YA) to high, medium and low trustworthy faces in whole brain, reward region, and FFA, whereas differentiation of the faces in the visual cortex was significant only for YA. Both LDA and linear SVM showed less accurate discrimination of facial category for OA than YA in reward region, and LDA showed less accurate discrimination for OA than YA in visual cortex, which suggested that reward region and visual cortex may be the brain regions that contribute to age-related dedifferentiation in impressions of face trustworthiness. Surprisingly, given previous evidence that OA show neural dedifferentiation in FFA responses to face vs. other categories, neither LDA nor linear SVM showed age differences in the discrimination of face trustworthiness in FFA.