Modeling the Brain as a Shannon Information Source for fMRI-Based Network Analysis in Early Alzheimers Disease Diagnosis

Alzheimers disease (AD) is an irreversible neurodegenerative disorder that gradually impairs memory, cognition, and behavior, making early diagnosis essential for slowing disease progression and improving patients quality of life. Functional Magnetic Resonance Imaging (fMRI) provides a noninvasive tool to study brain activity, yet many existing diagnostic models rely on black-box architectures that lack interpretability. In this study, we introduce a computational framework that models each anatomical brain region as a Shannon information source, thereby quantifying both the intrinsic information content of regions and the interactions among them. We used kernel density estimation to compute the probability density functions (PDFs) of voxel-level BOLD time series. From these PDFs, we derived regional entropy and pairwise Kullback-Leibler (KL) divergence measures. These measures were used to construct feature spaces representing information dynamics across the brain. We applied the framework to the ADNI resting-state fMRI dataset, which includes cognitively normal (CN), early mild cognitive impairment (EMCI), late mild cognitive impairment (LMCI), and AD subjects. Our findings indicate that entropy values increase with disease progression, while KL-based connectivity networks reveal a progressive loss of inter-regional interactions, especially in frontal, temporal, and parietal lobes. For classification, we trained multilayer perceptrons using voxel BOLD signals, entropy vectors, and KL divergence vectors. Models trained on KL features achieved the highest performance, outperforming both entropy-based and voxel-based approaches. These results demonstrate that the Shannon information source model offers an interpretable and statistically grounded approach for characterizing brain dynamics, while achieving superior diagnostic accuracy. Beyond AD, the proposed framework provides a generalizable tool for studying brain network alterations in neurological and psychiatric disorders.