Abstract

INTRODUCTION Verbal fluency tasks are widely used in primary progressive aphasia (PPA), but most studies rely only on total correct responses, overlooking qualitative features of the words produced. We applied a scalable computational framework to extract item-level features from fluency responses in semantic variant (svPPA) and logopenic variant PPA (lvPPA) to test their value for differential diagnosis.METHODS We analyzed animal fluency responses from 113 participants (40 svPPA, 40 lvPPA, 33 controls) using an automated pipeline extracting nine psycholinguistic features. Group differences were examined with (co)variance models, classification with logistic regression, and brain-behavior associations via structural magnetic resonance imaging.RESULTS All features except semantic variability distinguished svPPA from lvPPA. Models including features outperformed (area under the curve [AUC] = 0.86) those using only total correct or clinical variables (AUC = 0.60-0.68). Features related mainly to temporal lobe atrophy, whereas total correct also related to the angular gyrus.DISCUSSION Automated item-level analysis offers a sensitive, scalable method for supporting PPA diagnosis and monitoring.Highlights Automated item-level features from verbal fluency aid semantic variant primary progressive aphasia (PPA) versus logopenic variant PPA diagnosis. Item-level fluency features outperform total correct responses for classification. Item-level fluency features map onto syndrome-relevant temporal lobe atrophy. A scalable, fully automated pipeline enables integration into clinical practice. There is potential to support low-burden, objective monitoring of disease progression in PPA.