Search

Background: The classification of patients with primary progressive aphasia (PPA) into variants is time-consuming, costly, and requires combined expertise by clinical neurologists, neuropsychologists, speech pathologists, and radiologists., Objective: The aim of the present study is to determine whether acoustic and linguistic variables provide accurate classification of PPA patients into one of three variants: nonfluent PPA, semantic PPA, and logopenic PPA., Methods: In this paper, we present a machine learning model based on deep neural networks (DNN) for the subtyping of patients with PPA into three main variants, using combined acoustic and linguistic information elicited automatically via acoustic and linguistic analysis. The performance of the DNN was compared to the classification accuracy of Random Forests, Support Vector Machines, and Decision Trees, as well as to expert clinicians' classifications., Results: The DNN model outperformed the other machine learning models as well as expert clinicians' classifications with 80% classification accuracy. Importantly, 90% of patients with nfvPPA and 95% of patients with lvPPA was identified correctly, providing reliable subtyping of these patients into their corresponding PPA variants., Conclusion: We show that the combined speech and language markers from connected speech productions can inform variant subtyping in patients with PPA. The end-to-end automated machine learning approach we present can enable clinicians and researchers to provide an easy, quick, and inexpensive classification of patients with PPA.

People with primary progressive aphasia (PPA) present language difficulties that require lengthy assessments and follow-ups. Despite individual differences, people with PPA are often classified into three variants that present some distinctive language difficulties. We analyzed the data of 6 fluency tasks (i.e., "F", "A", "S", "Fruits", "Animals", "Vegetables"). We used random forests to pinpoint relevant word properties and error types in the classification of the three PPA variants, conditional inference trees to indicate how relevant variables may interact with one another and ANOVAs to cross-validate the results. Results indicate that total word count helps distinguish healthy individuals (N = 10) from people with PPA (N = 29). Furthermore, mean familiarity differentiates people with svPPA (N = 8) from people with lvPPA (N = 10) and nfvPPA (N = 11). No other word property or error type was relevant in the classification. These results relate to previous literature, as familiarity effects have been reported in people with svPPA in naming and spontaneous speech. Also, they strengthen the relevance of using familiarity to identify a specific group of people with PPA. This paper enhances our understanding of what determines word retrieval in people with PPA, complementing and extending data from naming studies.