Investigating Differences in Language Processing with Aphasia Disorder Using Graph Convolutional Networks

Abstract

Aphasia Affected Persons (AAPs) often get annoyed due to the dearth of a link between their internal monologue, which is also known as inner speech, and external words or overt speech. The main objective of this research is to classify the manners in which people with aphasia caused by stroke relate to themselves and others. This paper proposes a model employing a Graph Convolutional Network (GCN) aimed at achieving specific research objectives. Utilizing functional Magnetic Resonance Imaging (fMRI) data from 20 subjects, including individuals diagnosed with Aphasia Affected Persons (AAP) and Healthy Persons (HP), the study encompasses a balanced dataset of 11 females and 9 males. The methodology is designed to investigate elusive properties or relationships, thereby providing an enhanced understanding of the mechanisms underlying language processing in the brain. Data is pre-processed thoroughly using methods like Retrospective Image Correction (RETROICOR) and spatial smoothing is applied to eliminate physiological noise as well as enhance data quality. The GCN is trained on how patients judge words (JoW) and synonyms (JoS) so that it makes predictions about them after partitioning its data among different models. Evaluation of the proposed model’s performance included metrics such as patient accuracy and response time. As compared to age-matched controls, AAP shows greater inter-subject variations in brain activity between concrete and abstract words possibly due to an increased concreteness effect. In contrast, HPs reach their maximum accuracy in an abstract condition where it amounts to 99.65% at reaction time equaling 1981.33 milliseconds whereas AAPs whose control condition records 100% with reaction time around 1642.56 milliseconds. © (2024), (Greater Mekong Subregion Academic and Research Network). All Rights Reserved.