Exploring Aphasia Using Community Detection Analysis And Machine Learning

Aphasia, generally defined as an impairment in the production or comprehension of language, was first described in the late 19th century. Despite numerous critiques of the original model of aphasia proposed by Lichtheim in 1885 and advancements in neuroimaging, statistics and behavioral assessment techniques, the principles and sub-types Lichtheim proposed still play a large role in the sub-typing framework used in clinical aphasiology today. Therefore the current project sought to explore aphasia diagnoses using a data driven approach and modern statistical tools by examining the behavioral deficit profiles and their lesion correlates in a large cohort of patients. The project consisted of three studies. Study 1 sought to address concerns of poorly defined aphasia sub-types, by using community detection analysis (CDA) to cluster patients with aphasia based on their behavioral profiles and to then compare these clusters to the traditional sub-types to see how well they align. Building from Study 1, in the second study voxel-based lesion symptom mapping was used to identify neural regions specific to the clusters identified in Study 1. Finally, in Study 3 classifiers were built in attempts to try and predict a patients cluster based solely on their lesion profiles. Overall the results of the CDA and VLSM analyses did not align with the traditional model of aphasia and suggest that the primary distinction in aphasia after severity should be between phonological and semantic processing rather than production and comprehension as traditional models suggest. Further the classifier was successful in categorizing patients and/or predicting a patients deficit showing potential for tools such as these to be used clinics. In all the results of this study suggest that the field needs to start moving in a new direction in regards to the study and classification of aphasia types.Ph.D., Psychology -- Drexel University, 201