Data-driven spatiotemporal modeling reveals personalized trajectories of cortical atrophy in Alzheimer’s disease

Chunyan Li, Yutong Mao, Xiao Liu, Wenrui Hao · 2025

Read Data-driven spatiotemporal modeling reveals personalized trajectories of cortical atrophy in Alzheimer’s disease by Chunyan Li, Yutong Mao, Xiao Liu, Wenrui Hao (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-08847.txt.

Introduction

Chunyan Li, Yutong Mao, Xiao Liu, and Wenrui Hao model Alzheimer's disease as a moving system rather than a set of snapshots. Their framework builds a personal graph for each patient from MRI and PET scans, then learns that patient's own dynamical settings for how brain regions change together over time.

Applied to one thousand eight hundred ninety-one participants in a major Alzheimer's imaging program, the model tracks the joint evolution of amyloid, tau, neurodegeneration, and cognition. It predicts future paths better than standard clinical and imaging benchmarks. Each patient's settings separate distinct progression patterns and forecast cognitive decline better than standard biomarkers. Sensitivity analysis flags the regional network features that drive spread, matching the known weakness of temporolimbic and frontal areas. The authors offer the approach as a base for modeling disease one patient at a time.

Catalog record

CREATORChunyan Li, Yutong Mao, Xiao Liu, Wenrui Hao
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-08847

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Data-driven spatiotemporal modeling reveals personalized trajectories of cortical atrophy in Alzheimer’s disease Chunyan Li, Yutong Mao, Xiao Liu, Wenrui Hao Abstract Alzheimer’s disease (AD) is characterized by the progressive spread of pathology across brain networks, yet forecasting this cascade at the individual level remains challenging. We present a personalized graph-based dynamical model that captures the spatiotemporal evolution of cortical atrophy from longitudinal MRI and PET data. The approach constructs individualized brain graphs and learns the dynamics driving regional neurodegeneration. Applied to 1,891 participants from the Alzheimer’s Disease Neuroimaging Initiative, the model accurately predicts key AD biomarkers—including amyloid- β \beta , tau, neurodegeneration, and cognition—outperforming clinical and neuroimaging benchmarks. Patient-specific parameters reveal distinct progression subtypes and anticipate future cognitive decline more effectively than standard biomarkers. Sensitivity analysis highlights regional drivers of disease spread, reproducing known temporolimbic and frontal vulnerability patterns. This network-based digital-twin framework offers a quantitative, personalized paradigm for AD trajectory prediction, with implications for patient stratification, clinical trial design, and targeted therapeutic development. Keywords: Alzheimer’s disease, spatiotemporal mathematical model, Alzheimer’s Disease Neuroimaging Initiative (ADNI), amyloid- β \beta , tau, neurofunctional loss, sensitivity analysis 1 Introduction Alzheimer’s disease (AD) affects over 50 million people worldwide and represents the most common form of dementia, imposing a profound societal, economic, and personal burden [ 54 ] . Despite decades of intensive research and billions invested in therapeutic development, the underlying mechanisms that drive its onset and progression remain poorly understood, and no disease-modifying therapies have yet succeeded in the clinic. AD is characterized by extracellular amyloid-beta (A \beta ) plaques and intracellular neurofibrillary tangles composed of hyperphosphorylated tau, which disrupt synaptic function, promote neuronal loss, and manifest as progressive cognitive decline pathogenesis [ 24 , 2 ] . While the amyloid cascade hypothesis has provided a foundational framework, it fails to fully capture the multifactorial and dynamic nature of sporadic AD, which involves the interplay of amyloid and tau pathology, neuroinflammation, and other contributing biological processes. Recent advances in biomarker imaging have transformed our ability to study AD in vivo. Positron emission tomography (PET) allows visualization of A \beta and tau deposition, structural MRI quantifies cortical atrophy, and cerebrospinal fluid (CSF) assays provide complementary molecular insights [ 15 , 47 ] . Functional imaging further reveals network-level alterations linked to cognitive impairment [ 67 ] , emphasizing that the critical questions are not only what changes occur in AD, but how , where , and why pathology propagates across the brain. These data create an unprecedented opportunity to study AD as a spatiotemporally dynamic disease, but also highlight the limitations of conventional analytical approaches. Mathematical and computational modeling [ 38 , 31 , 34 , 12 , 33 , 63 , 3 , 26 ] has emerged as an essential tool to integrate multi-modal data, formalize mechanistic hypotheses, and generate testable predictions and therapeutic interventions in silico [ 21 , 6 , 46 , 58 , 14 ] . Ordinary differential equation (ODE) models have been used to describe A \beta production, aggregation, and clearance [ 21 , 6 , 8 ] , as well as tau hyperphosphorylation [ 65 , 7 ] . Extensions incorporating neuroinflammation and other modulatory factors have provided additional mechanistic insights [ 43 ] , and partial differential equation (PDE) frameworks capture the spatial propagation of pathology across brain regions [ 68 , 23 , 48 ] . Complementing these mechanistic approaches, machine learning and causal inference methods leverage large-scale clinical data to predict disease onset, stratify patients, and infer dynamic biomarker relationships [ 70 , 69 , 66 , 44 ] . Despite these advances, critical gaps remain. Most models focus on individual pathways or global biomarkers, failing to capture the integrated spatiotemporal dynamics of amyloid, tau, and neuroinflammation [ 51 , 52 , 59 , 11 , 61 ] . Few frameworks provide personalized, spatially resolved predictions from longitudinal multi-modal imaging, limiting their translational relevance. Existing approaches often rely on MRI, regionally aggregated CSF or PET measures, lacking the spatial granularity needed to identify critical regions for targeted intervention [ 35 , 64 , 50 ] . Here, we present a novel, mechanistic, data-driven framework to model the spatiotemporal progression of AD biomarkers at the individual level. Our approach formulates a system of PDEs on the brain’s functional connectivity network, capturing the regional propagation of four key AD biomarkers: A \beta , tau, cortical atrophy, and functional network alterations. The model incorporates subject-specific parameters inferred from PET and structural MRI data, enabling personalized predictions while preserving mechanistic interpretability. Comprehensive parameter inference and two-level sensitivity analyses identify critical disease drivers and brain regions. To our knowledge, this is the first mechanistic spatiotemporal model integrating multiple clinically validated biomarkers in a patient-specific framework. This framework provides three key advances: 1. It unifies temporal biomarker progression with spatial propagation across brain networks. 2. It enables patient-specific modeling of AD progression using multi-modal imaging biomarkers. 3. It facilitates the prediction of region-specific therapeutic responses, advancing precision medicine in AD. By bridging mechanistic modeling, graph-based brain networks, and longitudinal neuroimaging, our approach provides a comprehensive, personalized view of AD progression, with direct implications for early diagnosis, individualized prognosis, and targeted therapeutic strategies. 2 Results We applied the spatiotemporal generalized AD Biomarker Cascade (generalized ADBC) model to characterize individualized dynamics of clinically validated AD biomarkers and evaluate its potential for personalized disease forecasting. The model, governed by 14 parameters (6 global and 8 region-specific), captures heterogeneity across the brain’s functional network. Once estimated from neuroimaging data, it allows simulation of amyloid-beta (A \beta ), tau, neurodegeneration (N), and cognitive decline (C), while a two-level sensitivity analysis identifies region-specific vulnerabilities. The overall workflow is illustrated in Figure 1 , highlighting the model’s use for personalized biomarker prediction and sensitive-region identification. Figure 1 : Workflow integrating graph theory, neuroimaging biomarkers (A \beta -PET, tau-PET, cortical atrophy, and MMSE score), AD pathophysiology, and mathematical modeling to construct a personalized digital twin of AD progression. The model, formulated as a spatiotemporal system of PDEs, captures biomarker diffusion and nonlinear interactions on the brain’s functional connectivity network. Coupled with sensitivity analysis, the framework enables personalized forecasting and identification of region-specific vulnerabilities with sensitivity analysis. Individual-level simulations and predictions Figure 2 demonstrates the model’s performance for three representative individuals. The model achieves prediction accuracies of 96.15\% , 93.97\% , and 93.35\% for A \beta , tau, and N, respectively. Training fits (red rectangles), test evaluations (green), and future predictions (blue) are shown, with curves representing accumulated biomarker levels across 68 brain regions. By visualizing only two representative fitting points per biomarker, we simplify interpretation while preserving the spatial distribution of pathology. These results illustrate the model’s capability to forecast individualized biomarker trajectories, providing actionable information for clinicians to anticipate disease progression and personalize interventions. Figure 2 : Model simulations of three spatially dependent biomarkers (A \beta , tau, N) for three individuals. Training fits, test evaluations, and future predictions are color-coded as red, green, and blue, respectively. Regional curves show accumulated biomarker levels across 68 brain regions. Population-level evaluation To assess generalizability, we evaluated model performance across all subjects (Table 1 ). The nonhomogeneous (NH) model, which incorporates region-specific parameters, consistently outperforms the homogeneous (H) model in

Read the complete text