Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Autobiographical narratives from childhood and adolescence relate to brain structure in healthy older adults
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Kevin Statz1, Helena Balabin1,2, Laure Spruyt1, Bastiaan Tamm1,3, Stefan Sunaert1, Patrick Dupont1, Rik Vandenberghe1; 1Department of Neurosciences, Leuven Brain Institute, KU Leuven, Leuven, Belgium, 2Department of Computer Science, KU Leuven, Leuven, Belgium, 3Department of Electrical Engineering, KU Leuven, Leuven, Belgium
Introduction: Autobiographical memory requires a complex integration of multiple cognitive processes. We examined how variability in autobiographical narratives during healthy aging relates to individual differences in brain structure. Methods: We conducted the Autobiographical Interview in 64 cognitively unimpaired, amyloid-negative older adults (58–84 years; 36 female/28 male; CDR = 0; MMSE ≥ 27; neuropsychological performance within normal limits; Centiloid < 23.5) recruited from the Flemish Prevent AD Cohort (KU Leuven). The audio recordings of their autobiographical narratives were transcribed verbatim using a Dutch, transformer-based automatic speech recognition model. For each participant and each of the five life epochs, these transcripts were passed as input to a Dutch language model, RobBERT. Then mean pooling was applied to the resulting subword embeddings to derive an aggregated representation. Cortical thickness and hippocampal volumes were extracted from T1-weighted magnetic resonance imaging using FreeSurfer v7.4.1 and the Desikan-Killiany atlas. We used representational similarity analysis (RSA) to determine Spearman correlations between subject-wise representational dissimilarity matrices (cosine distance) of speech embeddings and brain morphometry, controlling for age, sex, education, and intracranial volume through partial correlation. Statistical significance was assessed through permutation testing (n=10,000). We used principal component analysis to identify components that correlated with narrative similarity and inspected their highest-loading regions descriptively. Results: Overall, inter-individual similarity in the full Autobiographical Interview correlated with similarity in cortical thickness (ρ = 0.210, p = 0.03), this remained significant when adding hippocampal volumes (ρ = 0.204, p = 0.046). We then analyzed each life epoch separately versus cortical thickness patterns. For the childhood autobiographical narratives, RSA yielded a Spearman correlation of ρ = 0.314 (p_corrected = 0.0075, Bonferroni-corrected across 5 life epochs). The RSA for the adolescence narratives also yielded a significant correlation, ρ = 0.222 (p_corrected = 0.0048). Early adulthood, midlife, and past-year narratives were not significant (ρ = 0.157, ρ = 0.064, ρ = 0.155 respectively; all p_corrected > 0.05). Across the five life epochs, RSA Spearman correlations decreased as memories got more recent (ρ = −0.9, p = 0.03), but this temporal gradient is only based on five levels and should be interpreted with caution. In the principal component analysis, we retained three components based on the scree plot elbow criterion that explained a cumulative 54% of the variance. The second principal component showed a significant RSA correlation with adolescence narratives (ρ = 0.154, p = 0.046). The 10 highest-loading regions were predominantly bilateral inferior frontal, orbitofrontal, and middle and inferior temporal cortices. Conclusion: In amyloid-negative, cognitively unimpaired older adults, inter-individual similarity in remote autobiographical narratives is associated with inter-individual similarity in frontotemporal cortical thickness patterns, particularly for regions supporting language control and semantic knowledge.
Topic Areas: Computational Approaches, Language Production