Posters

Neural Signatures of English Listening Comprehension: An fMRI Study Using Connectome-Based Predictive Modeling and Inter-Subject Similarity

As an essential skill to remain competitive in the global market, second-language listening proficiency plays an important role in modern students’ education, and has become an important focus of educational development. Given listening's key role in developing language ability, considerable research has been conducted in this field. Nevertheless, investigating neural mechanisms and signaling patterns in naturalistic contexts remains valuable due to their scarcity. In this study, we examined 48 participants' fMRI signals during the Test of English Listening Comprehension, which is currently used to assess high school students’ English performance and may influence their academic pathways. To capture the neural heterogeneity across different levels of listening ability, Connectome-based Predictive Modeling (CPM) and Inter-Subject Similarity (ISS) with moving-window were applied. The CPM results indicated that no reliable predictive connectome could be identified in the present datasets. On the other hand, ISS results, particularly in the fourth and fourteenth networks of Yeo et al.’s (2011) 17-network atlas, revealed that time points with higher similarity in the correct than in the incorrect responses differed across parts of the TELC, highlighting the substantial effect of test-item variability. These findings and the unbalanced distribution of English levels among participants (mostly with accuracy above 70%) may explain why a reliable CPM model could not be established.

Keywords: fMRI, Test of English Listening Comprehension (TELC), Connectome-based Predictive Modeling, Inter-Subject Similarity

A multi-contrast MRI brain template and CT skull model for Taiwanese macaque

Non-human primates (NHPs) have long been critical models in biomedical research. Compared to other lab animals, NHPs are phylogenetically closer to humans, and thus provide better models of the health and diseases in terms of genetics, anatomy, physiology and behavior. NHPs’ large brain, high intelligence and sociability make them especially suitable for the studies of higher cognitive functions and neuropsychiatric disorders. Taiwanese macaque (Macaca cyclopis) is the native primate living in Taiwan and is a close relative of the rhesus (Macaca mulatta) and Japanese macaques (Macaca fuscata). However, the feasibility of using it in biomedical research, especially in neuroscience, has rarely been studied. To facilitate this species to be used in brain research, a standard anatomical template is required for data analysis and comparison across subjects and studies. As a first and also critical step, we build an in-vivo MRI brain template with multiple MR contrasts, including T1W, T2W, FGATIR and DTI, collected from 16 Taiwanese macaques (11 from middle to elderly cohort, 5 from teenage cohort). In the meanwhile, an averaged skull model was derived from CT images and well-aligned to the MR brain template. These basic neuroimaging tools can be used for data visualization, surgical planning, parameter simulation for neuromodulation tools, comparative analysis among macaque species and so on.

Keywords: nonhuman primate, monkey, brain, atlas

Decoding the brain, or decoding the pipeline?  A four-pipeline multiverse re-analysis of a public visuomotor adaptation dataset

Decoding results are read as properties of neural representation, yet every step from preprocessing to classifier is an analyst's choice. How much of a map survives those choices is rarely quantified. We re-analysed OpenNeuro ds004562 (Song et al., 2023; n = 22). Two labs ran four pipelines on identical data: (A) the authors' own SPM/TDT toolchain; (B) the same analysis on fMRIPrep preprocessing; (C) a Python reimplementation (Nilearn, scikit-learn) on that preprocessing; (D) the partner lab's own pre-existing BrainVoyager analysis. At the time of this abstract submission, pipeline A and B are complete; C covers nine of eleven analyses; D is not yet in the comparison. Agreement is the voxelwise Spearman ρ between group maps in one mask. Nine contrasts have all three pipelines, and the ordering is identical in each: pipelines sharing preprocessing but differing downstream agree most (ρ = .87–.98), the pair differing only in preprocessing next (.66–.92), and the pair differing in both least (.63–.90). Preprocessing therefore dominates where decoding succeeds. The downstream implementation acts on a second axis: pipeline C decodes 0.2–1.5 points below B despite identical inputs—a level shift rather than a spatial rearrangement. Robustness is contrast-specific and invisible from one pipeline. Ongoing work adds the preparation-period analyses and pipeline D, including two analyses that were null in the original report, to ask whether a null result survives a change of pipeline.

Keywords: fMRI, Multi-Voxel Pattern Analysis, multiverse analysis, open dataset, reproducibility

Mapping Asymmetric Within-Network Connectivity in Resting-State fMRI Using Convergent Cross Mapping

Functional connectivity is conventionally estimated using symmetric measures such as Pearson correlation, which cannot represent interaction direction. This project explores convergent cross mapping (CCM), a state-space method for detecting coupling, as a complementary approach to mapping directional relationships in the human brain. Using resting-state fMRI from the open Human Connectome Project and Schaefer cortical parcellations, we construct participant-level directional connectivity maps within canonical resting-state networks. Preliminary results reveal asymmetric directional-detection maps derived from CCM cross-map-skill curves: for many reciprocal parcel pairs, one direction is detected more often across participants than its reverse. These findings suggest that CCM may reveal organization not represented by symmetric correlation. Potential applications include identifying directional network structure, characterizing influential parcels and pathways, and developing individualized connectivity features related to cognition, personality, emotional functioning, and mental health. By combining open neuroimaging data with a reproducible workflow, this project seeks to expand resting-state connectivity research from undirected association toward directional, nonlinear interaction.

Keywords: convergent cross mapping, functional connectivity, resting-state fMRI, Human Connectome Project, directional connectivity

Generalizability of Syntactic and Semantic Neural Representations: A Cross-Task fMRI MVPA replications of openneuro ds003604

This replication study investigates the stability and generalizability of syntactic and semantic representation developments in the human brain. Using functional magnetic resonance imaging (fMRI), we re-examined representational similarity and cross-task multi-voxel pattern analysis (MVPA) across target regions of interest (ROIs), including the inferior frontal gyrus (IFG), middle temporal gyrus (MTG), and superior temporal gyrus (STG), the critical analysis in Wang et al., (2020), derived from the openneuro dataset (ds003604). Our current implementation evaluates feature selection using a group-wise GLM contrast mask overlapped with AAL atlas definitions, alongside a parallel pipeline replicating the original individual-level GLM mask criteria. Preliminary findings from the group-wise mask reveal two key divergences from prior literature: (a) IFG Pattern Bimodality: Average correlation values (r) in the IFG exhibit weak positive tendencies rather than the expected negative correlations reported in published findings; (b) Cross-Task Generalizability: Pattern similarity across tasks within MTG and STG remains exceptionally high, matching or exceeding within-task correlations. These discrepancies highlight how spatial masking strategies and baseline noise calibration significantly alter representational metrics. Ongoing adjustments focus on individual-level localized masks to resolve signal directionality and validate cross-task neural decoding.

Keywords: developmental neuroscience, openneuro dataset, fMRI, MVPA, scaffolding hypothesis