Schedule of Events | Symposia

Sketchpad Series

Community Detection in Adults and Neonates

Poster Session D - Monday, March 31, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Nilanjan Chakraborty¹ (chakrabortyn@mst.edu), Ayoushman Bhattacharya², Jiaxin Tu², Donna Dierker², Soumen Lahiri², Adam Eggebrecht², Muriah Wheelock²; ¹Missouri University of Science and Technology, ²Washington University in Saint Louis

This poster aims at using the Weighted Stochastic Block Model algorithm to determine the optimal number of brain communities in neonates and compare the findings and benchmark the findings against other popular community detection algorithms. The findings and preliminary figures are included in the poster where we have obtained 15 communities in neonates. We used fMRI data from the Baby Connectome Project (BCP), which included an average of 16.9 minutes of MRI images from 301 full-term infants (8-30 weeks) acquired using a Siemens 3T Prisma scanner. The Weighted Stochastic Block Model (WSBM) algorithm was applied to identify communities. We bootstrapped the log-likelihood differences 2000 times to determine the optimal community number based on the 95% confidence interval and via a consensus algorithm. We applied WSBM, on the Baby Connectome Project (BCP) dataset and obtained K = 15 as the optimal number of communities using bootstrap Bayes factor intervals and identified community structure in the infant brain. Comparisons with Infomap on BCP dataset (Tu et al. 2024) showed moderately high Normalized Mutual Information values. Our preliminary results suggest that the number of communities in the infant brain is 15 and the community structure in the infant brains are not fully assortative. This is consistent with prior work using Infomap (Tu et al. 2024; Kardan et al. 2020; Eggebrecht et al. 2017) demonstrating higher order association networks are not fully formed in infancy.

Topic Area: METHODS: Neuroimaging