Modeling Spatial-Temporal Dependencies in Emotion-Related fMRI: A Nonparametric Bayesian Application to the NeuroEmo Dataset.

OBJECTIVES: To characterize the spatiotemporal brain activation patterns evoked by five culturally validated emotion categories-Calm, Afraid, Delighted, Depressed, and Excited-in an Indian sample, and to demonstrate the advantages of a nonparametric Bayesian general linear model (GLM) for naturalistic fMRI data. MATERIALS & METHODS: Functional MRI data were obtained from 40 healthy, right handed Indian adults (mean age 28.3 ± 9.14 years; 31 males, 9 females) via OpenNeuro (ds005700). Participants viewed 30 s film clips from the Affective Film Dataset from India, interleaved with white noise baselines. Data were preprocessed in SPM12, and regional time series were extracted from 90 cortical/subcortical AAL ROIs using MarsBaR 0.45. We applied the NPBayes fMRI toolbox to fit a spatiotemporal Bayesian GLM with a hierarchical Dirichlet process prior for subject clustering, and spatial basis-function regularization. Posterior inference used Variational Bayes, and activation was declared via posterior probability of inclusion (PPI) thresholded to control a 5% Bayesian false discovery rate. RESULTS: All emotion conditions engaged early and higher order visual cortices. Calm elicited focal lingual-cuneus activation; Afraid recruited middle/inferior temporal regions; Delighted and Excited amplified visual responses-with Excited also activating parietal attention networks; Depressed combined visual engagement with posterior cingulate/precuneus. The Bayesian framework revealed latent subject subgroups and provided threshold free, reproducible activation maps. CONCLUSION: Nonparametric Bayesian general linear model analysis of culturally relevant film stimuli yields nuanced insights into emotion-brain dynamics, controls Type I error without arbitrary thresholds, and uncovers interindividual heterogeneity, offering a robust tool for affective neuroimaging.