PSYC 42350 Final Project

This repository contains the code, analysis scripts, and results for a final project in the University of Chicago’s PSYC 42350 course. The project investigates how narrative content modulates neural synchrony using simultaneous EEG-fMRI data from an open-access dataset published by Telesford et al. (2023). Two main analytical approaches—inter-subject correlation (ISC) and representational similarity analysis (RSA)—were applied to examine whether coherent narratives evoke stronger neural alignment than non-narrative stimuli.

Data

I used an open-access dataset from Telesford et al. (2023), which includes EEG and fMRI recordings of participants viewing both narrative (e.g., Despicable Me) and non-narrative (e.g., Inscapes) video clips.
- Data Portal
- Direct Download Links

Research Questions

1. ISC: Do coherent narratives evoke stronger inter-subject correlation in fMRI compared to non-narrative stimuli?
   
2. Cross-Modal RSA: Does cross-modal EEG–fMRI representational similarity differ between narrative and non-narrative conditions, and can these multimodal signals classify whether a segment is narrative vs. non-narrative?

Methods

1. Dataset Preprocessing:

   • 22 healthy adults (ages 23–51) with no history of psychiatric/neurological illness.
     
   • EEG recorded at ~5000 Hz, fMRI at 3T with TR=2.1 s.
     
   • Preprocessing (motion correction, MNI alignment, etc.) done with the Connectome Computation System (CCS).
     

2. Segmentation & Analyses:

   • Segmentation: Both EEG and fMRI time series were segmented into 10-second windows, creating averaged neural patterns per segment.
     
   • ISC: Only fMRI data were used, correlating each subject’s voxel-wise time course with the mean of all other subjects. ISC maps were aggregated within the 17 Yeo networks.
     
   • RSA: Constructed representational dissimilarity matrices (RDMs) for EEG and fMRI (using 1 – Pearson correlation) and computed cross-modal alignment.
     
   • Classification: A linear SVM was used to classify narrative vs. non-narrative segments based on RSA features.

Results

1. ISC (fMRI):

   • Narratives induced significantly higher inter-subject correlation across 16 of the 17 Yeo networks, particularly in high-level cortical networks such as the Control/Frontoparietal and Default Mode networks.
     
   • A repeated-measures ANOVA confirmed a robust effect of network on the ISC difference (Narrative minus Non-Narrative).

2. Cross-Modal RSA (EEG–fMRI):

   • Group-level RSA showed a slight positive trend for narratives (r = 0.003) vs. non-narratives (r = –0.009), but the difference was not statistically significant (paired t-test p = 0.148).
     
   • Classification accuracy using RSA features was 52.8%, only marginally above chance.
     

These findings indicate that while narratives strongly synchronize brain activity across individuals in fMRI, the cross-modal representational alignment (EEG vs. fMRI) remains modest under current analytic parameters.

Virtual Environment

To replicate this analysis, you can use the provided requirements.txt in a Python 3.11 environment:

```bash

Create the virtual environment:

python3.11 -m venv venv

Activate the virtual environment:

source venv/bin/activate

Install required packages:

python3 -m pip install -r requirements.txt