Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Unveiling Accelerated Brain Aging through Discourse Analysis: Leveraging Natural Language Processing and Neuroimaging
Poster A7 in Poster Session A, Tuesday, October 24, 10:15 am - 12:00 pm CEST, Espace Vieux-Port
Ezequiel Gleichgerrcht1, Haris Rashid1, Ronak Pipaliya2, Rebecca Roth1, Federico Rodriguez Porcel2, Samaneh Nemati3, Sarah C. Wilson3, Sarah E. Newman-Norlund3, Julius Fridriksson3, Leonardo Bonilha1; 1Emory University, 2Medical Unviersity of South Carolina, 3University of South Carolina
Identifying non-invasive biomarkers of accelerated brain aging holds significant potential for early detection and intervention in neurodegenerative disorders. Complex cognitive processes, such as discourse, are promising indicators of brain health, often manifesting with subtle deficits years before other neuropsychological impairments become apparent. However, traditional discourse analysis is time-consuming and requires highly trained examiners, posing challenges for larger-scale population studies. Natural Language Processing (NLP) offers a solution to this problem by using machine learning approaches to derive discourse features in a systematic and automated way. In this study, we applied NLP to quantify lexical features generated from verbatim transcriptions of responses to the "Cookie Theft" picture from 210 healthy participants in the Aging Brain Cohort (ABC@USC). Participants underwent 3T brain MRI, including T1-weighted and diffusion tensor imaging (DTI). A machine learning model that trained on a very larger cohort of healthy participants (BrainAgeR, n=3377) was applied to the T1-weighted images to estimate an imaging-derived brain age. The brain age gap (BAG), defined as the difference between brain age and chronological age, was computed as a measure of brain health, with higher BAG values indicating premature brain aging (brain age > chronological age). Additionally, an individualized structural connectome was computed from DTI using the Johns Hopkins University parcellation atlas. Betweenness centrality (BC), a graph theory measure reflecting network hubness, was computed for each brain region and averaged across the left-hemisphere language network as well as the whole brain. Our findings revealed that chronologically older participants exhibited a greater variety of words (r = 0.42, p = 0.04), higher lexical sophistication (r = 0.42, p < .001), and increased lexical diversity (r = 0.1, p = 0.01). However, across all ages, participants with premature brain aging exhibited fewer different words (r = -0.16, p = 0.02) along with decreased lexical sophistication (r = -0.19, p = 0.007), more narrow lexical word variation (r = -0.14, p = 0.039), and diminished lexical diversity (r = -0.19, p = 0.007). In turn, premature brain aging was associated with increased global average hubness (r = 0.22, p < .001), but there was a significant decrease in the average hubness of the left-hemisphere language network (r = -0.15, p = 0.03). In summary, we leveraged NLP to show accelerated brain aging is associated with decreased lexical complexity and diversity in a large cohort of healthy individuals. Premature brain aging was also associated with loss of average hubness specifically in the language network. Understanding how age-related brain structural changes affect discourse may offer a unique opportunity to develop non-invasive biomarkers of advanced brain aging and identify earlier stages of cognitive decline.
Topic Areas: Language Production, Methods