alzheimer-pet-suvr-network-models

# High-fidelity and Network-based Spatio-temporal Mathematical Models of Alzheimer's Disease Progression > A novel framework comparing 3D patient-specific brain geometries with reduced network-based connectome models for predicting amyloid-beta and tau protein propagation in Alzheimer's disease. ## Metadata - **Source**: arXiv:2604.18470v1 - **Authors**: Beatrice Caon, Mattia Corti, Francesca Bonizzoni, Paola F. Antonietti - **Published**: 2026-04-20 - **Category**: Computational Neuroscience, Neurodegeneration ## Core Methodology ### Problem Statement Alzheimer's disease (AD) progression involves the misfolding and accumulation of two toxic proteins: - **Amyloid-beta (Aβ)** plaques - **Tau** neurofibrillary tangles Mathematical models provide quantitative tools for monitoring disease progression and understanding the spatio-temporal dynamics of protein propagation. ### Dual-Approach Framework #### Approach 1: High-Fidelity 3D Biophysical Model - **Geometry**: Patient-specific 3D brain geometries reconstructed from MRI - **Governing Equations**: Reaction-diffusion PDEs on complex geometries - **Advantages**: Most accurate and biologically consistent description - **Limitations**: Computationally demanding #### Approach 2: Reduced Network-Based Model - **Graph Structure**: Brain connectome as a graph (nodes = regions, edges = white matter tracts) - **Formulation**: Graph Laplacian-based dynamics - **Advantages**: Cheaper computational cost - **Limitations**: Not always able to achieve reliable results across all brain regions ### Mathematical Formulation #### 3D Model ``` ∂u/∂t = D∇²u + R(u) (Reaction-diffusion equation) ``` Where: - u = protein concentration (Aβ or tau) - D = diffusion coefficient - R(u) = reaction term (protein production/clearance) #### Network Model ``` du/dt = -L_G · u + R(u) (Graph Laplacian dynamics) ``` Where: - L_G = graph Laplacian of the brain connectome - u = protein concentration at each node - R(u) = reaction term ### Validation Strategy - **PET Tracers**: 18F-AZD4694 (amyloid), 18F-MK6240 (tau) - **Data Type**: PET Standardized Uptake Value Ratios (SUVR) - **Comparison**: Model predictions vs. clinical PET-SUVR data - **Sensitivity Analysis**: Quantify parameter influence on concentration patterns ## Implementation Guide ### Prerequisites - **MRI Processing**: FreeSurfer or similar for brain geometry reconstruction - **Numerical PDE Solvers**: FEniCS, COMSOL, or custom finite element code - **Connectome Data**: Diffusion MRI tractography (e.g., from HCP, ADNI) - **PET Analysis**: SUVR calculation pipelines ### Step-by-Step #### Step 1: Data Preparation 1. Acquire structural MRI (T1-weighted) 2. Segment brain into regions of interest 3. Reconstruct 3D surface/volume meshes 4. Process diffusion MRI for tractography (connectome) 5. Acquire PET images and calculate SUVR maps #### Step 2: Model Setup (3D) ```python # Pseudo-code for 3D model setup import fenics as fn # Load brain geometry mesh = fn.Mesh('brain_geometry.xml') V = fn.FunctionSpace(mesh, 'P', 1) # Define reaction-diffusion problem u = fn.Function(V) v = fn.TestFunction(V) # Diffusion term D = 0.1 # diffusion coefficient diffusion = D * fn.dot(fn.grad(u), fn.grad(v)) * fn.dx # Reaction term (example: logistic growth + clearance) alpha = 0.5 # production rate beta = 0.3 # clearance rate reaction = (alpha * u * (1 - u) - beta * u) * v * fn.dx # Time stepping F = (u - u_n)/dt * v * fn.dx + diffusion - reaction ``` #### Step 3: Model Setup (Network) ```python import numpy as np import networkx as nx from scipy.sparse import csr_matrix from scipy.sparse.linalg import expm_multiply # Load connectome connectome = np.load('brain_connectome.npy') # N x N connectivity matrix G = nx.from_numpy_array(connectome) L = nx.laplacian_matrix(G) # Graph Laplacian # Network reaction-diffusion N = len(connectome) u = np.zeros(N) # Initial protein concentration # Simulation loop for t in range(n_steps): # Graph Laplacian diffusion + reaction dudt = -L.dot(u) + reaction_term(u) u = u + dt * dudt ``` #### Step 4: Parameter Estimation - Use sensitivity analysis to identify influential parameters - Calibrate against PET-SUVR data - Compare regional SUVR predictions #### Step 5: Model Validation - Calculate prediction error vs. clinical data - Compare 3D vs. network model performance - Assess biological plausibility ### Code Example: Sensitivity Analysis ```python def sensitivity_analysis(model_func, params, param_ranges): """ Perform sensitivity analysis for model parameters. Args: model_func: Function that runs the model params: Dictionary of parameter values param_ranges: Dict of {param_name: (min, max)} Returns: sensitivity_scores: Dict of parameter importance """ from SALib.sample import saltelli from SALib.analyze import sobol problem = { 'num_vars': len(param_ranges), 'names': list(param_ranges.keys()), 'bounds': list(param_ranges.values()) } param_values = saltelli.sample(problem, 1024) outputs = [] for params_sample in param_values: params_dict = dict(zip(param_ranges.keys(), params_sample)) output = model_func(**params_dict) outputs.append(output) Si = sobol.analyze(problem, np.array(outputs)) return Si # Example usage param_ranges = { 'D': [0.01, 1.0], # Diffusion coefficient 'alpha': [0.1, 1.0], # Production rate 'beta': [0.01, 0.5], # Clearance rate 'u0': [0.0, 0.5] # Initial concentration } ``` ## Applications ### Clinical Applications - **Disease Progression Prediction**: Forecast tau/Aβ spread across brain regions - **Treatment Planning**: Identify optimal intervention targets - **Clinical Trial Design**: Stratify patients by predicted progression rate - **Biomarker Development**: Identify early detection signatures ### Research Applications - **Pathology Understanding**: Mechanistic insights into protein propagation - **Model Comparison**: Evaluate trade-offs between accuracy and computational cost - **Connectomics**: Study role of network topology in disease spread - **Cross-Disease Analysis**: Apply to other proteinopathies (Parkinson's, CTE) ## Pitfalls ### Model Limitations - **3D Model**: Computationally expensive for large-scale studies - **Network Model**: May miss local heterogeneity within regions - **Parameter Identifiability**: Multiple parameter combinations may fit data equally well - **Patient Variability**: Single model may not capture all patient trajectories ### Validation Challenges - **PET Noise**: SUVR measurements have inherent uncertainty - **Regional Variability**: Different brain regions may require different parameters - **Longitudinal Data**: Limited availability of multi-timepoint data ### Implementation Notes - **Mesh Quality**: Poor mesh quality can destabilize 3D simulations - **Graph Construction**: Connectome quality strongly affects network model results - **Boundary Conditions**: Careful handling of brain boundaries required ## Related Skills - brain-connectivity-analysis - graph-laplacian-denoising - brain-network-controllability - brain-dit-fmri-foundation-model - computational-lesions-multilingual-language-models ## References - Caon et al. (2026). High-fidelity and Network-based Spatio-temporal Mathematical Models of Alzheimer's Disease Progression. arXiv:2604.18470v1 - ADNI (Alzheimer's Disease Neuroimaging Initiative): adni.loni.usc.edu - Raj et al. (2012). Network diffusion model of disease progression. Neuron. - Fornari et al. (2019). Practicalities of graph-based models for Alzheimer's disease.