2026 ECNP Congress

PS02-2159

Signals versus synchrony: decoding affective processing from voxel activation versus functional connectivity in task-based functional magnetic resonance imaging

Cécile Abati — M. Marxen, D. Grotegerd, L. Esser, K. Hierl, U. Dannlowski, R. Zahn, P. Kanske, R. Leenings, K. Förster · Technische Universität Dresden, Clinical Psychology and Behavioral Neuroscience Faculty of Psychology, Dresden, Germany

2026 ECNP Congress

Introduction: Emotions emerge from distributed patterns of brain activity, yet it remains debated whether voxel activation (VA) patterns or functional connectivity (FC) between brain regions provide a better basis for decoding affective states. Different emotions may recruit overlapping brain regions at similar activation levels, making them difficult to distinguish via VA alone, whereas FC captures inter-regional temporal dependencies that may encode additional emotion-discriminative information. Two prior studies [1], [2] provided initial evidence that FC outperforms VA, but both studies were constrained by small sample sizes (N = 38 and N = 12) and relied on single classifiers. This preregistered study provides the first well-powered comparison of VA-based and FC-based decoding of affective processing in a large sample, testing the hypothesis that FC outperforms VA across multiple levels of emotional granularity.

Methods: We analysed task-based fMRI data from 389 healthy participants from the Münster Neuroimaging Cohort (MNC) who performed an emotional face-matching task. The task comprises four face conditions using NimStim stimuli, happy, fearful, angry, neutral, and a shape-matching control. Both whole-brain and emotion-specific parcellations are defined using the Brainnetome atlas, with 36 emotion-related ROIs derived from a meta-analysis of 105 fMRI studies on emotional face processing [3]. For VA, condition-level beta maps are extracted from a standard GLM; for FC, pairwise Pearson correlations between ROI time series are computed, yielding one beta map and one FC matrix per condition per subject (1,945 each). Our machine-learning pipeline, implemented in PHOTONAI [4], employs nested cross-validation with hyperparameter optimisation, evaluating classifiers (SVC and decision tree) through a switch approach ensuring independent pipeline optimisation for each feature type. Classification follows a hierarchical strategy: face vs. shape (H1), emotional vs. neutral face (H2), and specific emotion categories (H3), with VA and FC compared using one-tailed paired permutation tests (α = 0.05). For further details, see OSF: https://osf.io/gub38/overview? view_only=77bf180f6e784c1493d0905a083ab102

Expected Results: Based on prior studies [1], [2], we expect both VA and FC to classify faces vs. shapes and emotional vs. neutral faces above chance. Critically, we predict that FC will outperform VA at each hierarchical level, consistent with the view that inter-regional coordination captures richer emotion-discriminative information than local activation alone. At the specific emotion level, we anticipate lower overall accuracy given overlapping neural substrates but expect the FC advantage to be maintained.

Conclusion: This preregistered study provides the first well-powered, rigorously cross-validated comparison of VA-based and FC-based decoding of affective processing. If FC outperforms VA, this would support network-based models in which emotions are primarily represented through inter-regional dynamics rather than local activation. Importantly, this study establishes the methodological foundation for applying the same decoding framework to a clinical sample of patients with major depressive disorder of the MNC cohort, where disrupted emotion processing and altered functional connectivity have been consistently reported. By clarifying how affective states are encoded in the healthy brain, this work informs the development of connectivity-based neuroimaging markers for affective disorders.

References

  1. S. P. Pantazatos, A. Talati, P. Pavlidis, and J. Hirsch, “Decoding Unattended Fearful Faces with Whole-Brain Correlations: An Approach to Identify Condition-Dependent Large-Scale Functional Connectivity,” PLOS Comput. Biol., vol. 8, no. 3, p. e1002441, Mar. 2012, doi: 10.1371/journal.pcbi.1002441.
  2. C. Liu, Y. Wang, X. Sun, Y. Wang, and F. Fang, “Decoding six basic emotions from brain functional connectivity patterns,” Sci. China Life Sci., vol. 66, no. 4, pp. 835–847, Apr. 2023, doi: 10.1007/s11427-022-2206-3.
  3. P. Fusar-Poli et al., “Functional atlas of emotional faces processing: a voxel-based meta-analysis of 105 functional magnetic resonance imaging studies,” J. Psychiatry Neurosci. JPN, vol. 34, no. 6, pp. 418–432, Nov. 2009.
  4. R. Leenings et al., “PHOTONAI—A Python API for rapid machine learning model development,” PLOS ONE, vol. 16, no. 7, p. e0254062, Jul. 2021, doi: 10.1371/journal.pone.0254062.