Early Detection of Alzheimer’s Disease Using a Shannon Information Source Model of the Brain

Date

2026-1-23

Author

Şekerci, Esra

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

45
views

38
downloads

Cite This

Alzheimer’s disease is associated with progressive disruption of resting-state functional organization that is difficult to characterize reliably using correlation- and graph-based summaries alone. This study evaluates an information-theoretic representation of rs-fMRI that models each anatomical region as an information source and quantifies both intra-regional uncertainty (Shannon entropy) and inter-regional distributional dissimilarity (Kullback–Leibler divergence). Using ADNI-2 rs-fMRI, we curated a four-class cohort (CN, EMCI, LMCI, AD) comprising 125 subjects and 424 sessions, with all sessions standardized to 140 volumes. After SPM-based preprocessing and AAL parcellation (90 ROIs), we derived three session-level feature families: BOLD, Entropy, and KL. On a stratified 90/10 train–test split with cost-sensitive learning and cross-validated hyperparameter search, information-theoretic features markedly improved discrimination, achieving test balanced accuracy of 0.9643 for Entropy and 0.9643 for KL (macro-F1: 0.9419 and 0.9514), compared to 0.8798 for BOLD. Errors concentrated in the transitional MCI stages, while CN and AD were consistently classified without error. These findings indicate that entropy and KL divergence capture disease-relevant distributional alterations in rs-fMRI that enhance stage-wise classification across the AD continuum.

Subject Keywords

Alzheimer’s Disease, fMRI, Brain Networks, Entropy, Kullback-Leibler Divergence, Multi-Class Classification.

URI

https://hdl.handle.net/11511/118513

Collections

Graduate School of Informatics, Term Project