Leveraging language and cognitive data for PPA subtyping: A systematic review of AI-based approaches

Primary Progressive Aphasia (PPA) is a neurodegenerative disorder marked by a gradual and selective decline in language. Accurate classification into its three clinical variants—nonfluent/agrammatic (nfvPPA), semantic (svPPA), and logopenic (lvPPA)—is essential but often limited by the time demands and expertise required for traditional assessments. This systematic review evaluates the application of artificial intelligence (AI) in the detection and classification of PPA variants using language and cognitive data. Following PRISMA 2020 guidelines, 14 peer-reviewed studies published between 2014 and 2024 were included. Studies were grouped by input modality: transcribed speech, acoustic features, multimodal data, and language-focused neuropsychological or task-based inputs (excluding studies based solely on general cognitive screening tools). Each was analyzed for methodological approach, AI technique, classification performance, and clinical relevance. AI-based approaches demonstrated high accuracy in distinguishing PPA variants. Transcribed linguistic features provided a practical and effective input source, while acoustic features were particularly sensitive to motor speech deficits in nfvPPA. Multimodal methods achieved the highest classification performance, and task-based models relying on language-oriented standardized assessments yielded interpretable and clinically applicable results. AI-driven analysis of language and cognitive data shows strong potential for improving PPA diagnosis and subtype classification. Future work should address limitations such as methodological variability, and lack of pathological validation. Advancements in cross-linguistic datasets, model transparency, and clinical integration will be essential for broader adoption.