Lindsay, H., Tröger, J., Alexandersson, J., & König, A. (2020). What Difference Does It Make? Early Dementia Detection Using the Semantic and Phonemic Verbal Fluency Task. Resources and ProcessIng of Linguistic, Para-Linguistic and Extra-Linguistic Data from People with Various Forms of Cognitive/Psychiatric/Developmental Impairments (RaPID-3), 46–82.
Verbal Fluency (VF) tasks are common cognitive tests that are used in the diagnosis of early stages of Dementia. There are two main types of VF tasks; Semantic Verbal Fluency (SVF) and Phonemic Verbal Fluency (PVF). While much work has been done on automatic diagnostic relevance of the SVF, research on the automatic analysis of the PVF task or a combination of both remains minimal. This paper explores methods of extracting features from the SVF and the PVF task according to clinical and temporal methods, as well as how combined within-subject features from both tasks can increment classification performance. We investigate an early diagnostic scenario with a binary classification between healthy controls (N=8) and those with mild cognitive impairment (N=19), a likely precursor to dementia. Synthetic data augmentation (SMOTE) is used to balance the data set and multiple machine learning models— logistic regression, support vector machines with linear and radial basis function, and a multi-layer perceptron—are used to evaluate the features. The best performance comes from combining SVF, PVF and novel joint within-subject features (AUC > 0.90) for multiple machine learning methods.