Multimodal Fusion Analysis of [18F]Florbetapir PET and Multiscale Functional Network Connectivity in Alzheimer’s Disease

Bikesh Bimali, Nigar Khasayeva, Adithya Ram Ballem, Geethanjali Nagaboina, Jiayu Chen, Alex Fedorov, James J Lah, Allan Levey, Vince Calhoun, Armin Iraji

Confirmation: I have read and agree with the IEEE BHI 2025 conference submission's policy on behalf of myself and my co-authors.

Keywords: Multimodal fusion, Alzheimer’s disease, Independent Component Analysis, Amyloid-beta, Positron Emission Tomography, Resting-state fMRI, Functional Network Connectivity, APOE genotype

TL;DR: We fused amyloid-PET with fMRI-derived connectivity to reveal how Alzheimer’s disease links brain chemistry and network function, uncovering joint patterns of amyloid buildup and connectivity disruption often missed by single-modality studies.

Abstract:

Accumulation of amyloid-beta plaques and disrup-
tion of intrinsic brain networks are two important character-
istics of Alzheimer’s disease (AD), yet the relationship between
amyloid accumulation and network dysfunction remains unclear.
In this study, we integrated [18F]Florbetapir PET and resting-
state fMRI (rsfMRI) derived Functional Network Connectivity
(FNC) from 552 temporally matched longitudinal PET–rsfMRI
sessions across 395 participants spanning Cognitively Normal
(CN), Mild Cognitive Impairment (MCI), and Dementia stages.
With a model order of 11, joint Independent Component Analysis
(jICA) was applied to the fused PET–FNC data, identifying 11
stable components, of which 9 PET-derived components corre-
sponded to previously characterized brain regions or networks.
The multimodal analysis revealed disease progression markers,
including (1) a pattern of reduced subject loadings across clinical
stages (CN > MCI > Dementia) in white matter and cerebellar
regions, reflecting structural degeneration; (2) increased amyloid
accumulation in affected individuals in grey matter regions,
particularly in frontal, sensorimotor, extended hippocampal, and
default mode network (DMN) regions, accompanied by functional
connectivity alterations that reflected both compensatory and
disruptive network dynamics. We identified PET-derived com-
ponents that captured distinct stages of disease progression, with
the DMN component emerging as a late-stage biomarker and a
white matter component showing early-stage changes with limited
progression thereafter. Additionally, several components showed
significant variation in loadings between APOE ε4 carriers
and non-carriers, linking the multimodal signatures to a well-
established genetic risk factor for AD.

Track: 3. Imaging Informatics

Registration Id: 4JNV9YY8L3B

Submission Number: 141