Brain Fingerprinting-Based Community Structure

• Published in: Proceedings (International Symposium on Biomedical Imaging) pp. 1 - 4
• Main Authors: Athamnah, Sema, Aviyente, Selin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.04.2025
• Subjects: Biomedical imaging; Brain; community detection; Feature extraction; Fingerprint recognition; fingerprinting; Functional connectivity; Functional magnetic resonance imaging; multilayer graphs; Nonhomogeneous media; Optimization; Precision medicine; Principal component analysis; signed graph; Vectors
• DOI: 10.1109/ISBI60581.2025.10981160

The ability to uniquely characterize individual subjects based on their functional connectome (FC), i.e., fingerprinting, is key for progress toward precision medicine. Over the past few years, a variety of methods have been proposed for fingerprinting. While early approaches focused on using either the whole brain connec-tivity or pre-defined subnetwork connectivity to discriminate between subjects, more recently different feature extraction methods such as principal component analysis (PCA) have been used to ex-tract discriminative features. However, these methods do not iden-tify functionally interpretable network components, i.e., commu-nities, that contribute most to discrimination across subjects. In this paper, we introduce a fingerprinting approach based on community structure. In particular, we introduce a signed multilayer modularity optimization approach to identify the community structure across a group of subjects from resting state fMRI FCs. The optimal community structure for identifying subjects and discrimi-nating between different tasks is learned through a community in-dicator vector which quantifies the relevance of each community for a subject. The proposed framework is tested on the Midnight Scan Club (MSC) data across different tasks. High identification rates between resting state and different tasks indicate that an indi-vidual's community indicator vector is intrinsic to that individual independent of the task. The proposed framework also provides interpretability to fingerprinting by determining the most discrimi-native brain regions between task pairs.