Classification of primary progressive aphasia: Do unsupervised data mining methods support a logopenic variant?

Area of Science:

• Neuroscience
• Computational Linguistics
• Data Science

Background:

• Primary progressive aphasia (PPA) is a neurodegenerative disorder affecting language.
• Current diagnostic models propose distinct PPA variants, including logopenic PPA.
• The validity of these distinct variants requires further investigation.

Purpose of the Study:

• To evaluate whether data mining techniques support a three-group diagnostic model of PPA.
• To compare the three-group model against a two-group model using unsupervised learning.
• To determine if logopenic PPA can be clearly distinguished as a separate variant.

Main Methods:

• Applied unsupervised learning methods (Expectation Maximization, K-Means, X-Means, Hierarchical Clustering, Consensus Clustering) to a dataset of 155 PPA patients.
• Utilized demographic, clinical, and neuropsychological attributes for clustering.
• Analyzed patient data using various algorithms and attribute sets.

Main Results:

• Unsupervised learning consistently identified two main patient groups across all analyses.
• One group comprised predominantly agrammatic/non-fluent and some logopenic cases.
• The other group mainly consisted of semantic and logopenic cases; no distinct logopenic PPA group emerged.

Conclusions:

• Data mining approaches using unsupervised learning do not support the distinction of logopenic PPA as a separate variant.
• The findings suggest a potential overlap or integration of logopenic cases within broader PPA classifications.
• Further research may refine PPA diagnostic criteria based on data-driven approaches.