Application of the dual stream model to neurodegenerative disease: Evidence from a multivariate classification tool in primary progressive aphasia.
Lynsey M Keator, Grigori Yourganov, Andreia V Faria and 2 others
PMID 35493273WHAT IT FOUND
Machine learning accurately classified PPA variants using brain scans and language tests, supporting the dual stream model.
Semantic deficits and ventral atrophy identified svPPA, while repetition deficits and dorsal atrophy identified lvPPA and nfaPPA, though distinguishing between these two dorsal variants remained difficult.
Key findings
01Cortical volume measurements allowed accurate classification of all three PPA variants, with the highest accuracy for distinguishing nonfluent/agrammatic from semantic variants.
02Behavioral classification using language scores successfully distinguished semantic variant PPA from the other two variants, but failed to distinguish between logopenic and nonfluent/agrammatic variants.
03The Pyramids and Palm Trees Test was the single most effective behavioral measure for identifying semantic variant PPA.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The sample size for Diffusion Tensor Imaging (white matter) analysis was significantly smaller (n=48) than the full cohort, reducing the robustness of those findings. The study did not include apraxia of speech scores as a primary input for the behavioral classifier, which may have hindered the ability to distinguish between logopenic and nonfluent/agrammatic variants. Disease severity was not controlled for in the analysis, and there is no consensus on a single measure for severity in PPA. The regions of interest were constrained to the dual stream model, potentially missing other relevant brain areas involved in language processing.
Declared interests
Argye E. Hillis receives compensation from the American Heart Association for editorial activities and from Elsevier for editorial activities for Practice Update Neurology.
The easy way to misread this
Do not assume this machine learning model is ready for clinical diagnosis. The study used a highly specialized research battery and automated segmentation methods not typically available in standard clinical settings. Furthermore, the model failed to distinguish between logopenic and nonfluent/agrammatic variants using behavioral data alone, meaning it cannot replace comprehensive clinical evaluation for these subtypes.