| name | from-read-out-geometry-to-in-silico-stimulation |
| description | Distributed functional-connectivity signature of Alzheimer's disease methodology using subject-specific reservoir-computing models to reconstruct individual lagged functional connectivity and develop personalized neuromodulation strategies. Shows that optimal stimulation targets are distributed patterns rather than focal sites, requiring model-informed targeting based on therapeutic responsiveness rather than read-out deviation magnitude. |
| metadata | {"arxiv_id":"2607.24356","published":"2026-07-27","authors":"Cristiano Capone, Enza Cece, Andrea Ciardiello, Guido Gigante, Evaristo Cisbani, Maurizio Mattia","tags":["alzheimer-disease","functional-connectivity","reservoir-computing","neuromodulation","personalized-medicine","brain-networks","computational-neuroscience"]} |
| license | Complete terms in LICENSE.txt |
From Read-Out Geometry to In-Silico Stimulation
Overview
This methodology addresses a critical question in Alzheimer's disease (AD) treatment: whether the functional connectivity (FC) signature reduces to focal sites or requires distributed network intervention. Using subject-specific reservoir-computing models, the approach reconstructs individual lagged FC and develops personalized neuromodulation strategies that account for network-level therapeutic responsiveness.
Key Contributions
1. Subject-Specific Reservoir Computing Models
- Cross-subject-identifiable models: Each individual's model can be mapped to a common template
- Lagged FC reconstruction: Models accurately reconstruct temporal dynamics of resting-state FC
- Functional vs Structural Read-outs: Functional read-out prioritized over structural atrophy because it's modifiable through stimulation
2. Distributed Connectivity Signature
- Ideal correction: Mapping patient dynamics onto control template requires distributed changes in connectivity kernel
- Coordinated multi-site patterns: Not reducible to single focal targets
- Therapeutic responsiveness: Optimal targets identified by effect on disease discriminant, not by magnitude of deviation
3. Personalized Neuromodulation Strategies
- Single-site drive failure: Stimulating node with largest kernel change fails even at supra-physiological amplitudes
- Effect-based targeting: Selecting sites by their effect on disease discriminant achieves complete individualized reclassification
- Real-time closed-loop control: Comparable efficacy at lower dose using only causally available information
4. Clinical Implications
- Cortical and heterogeneous targets: Optimal stimulation sites vary across patients
- Model-informed targeting: Site selection must be based on network therapeutic responsiveness
- Personalized medicine: One-size-fits-all approaches are insufficient for AD neuromodulation
Methodology
Model Architecture
class SubjectSpecificReservoir:
def __init__(self, num_nodes, connectivity_kernel):
self.reservoir = ReservoirNetwork(num_nodes)
self.connectivity_kernel = connectivity_kernel
self.readout_weights = None
def fit_to_subject_fc(self, subject_fc_data):
self.reservoir.train_dynamics(subject_fc_data)
self.readout_weights = self._train_classification_readout()
return self
def compute_ideal_correction(self, control_template):
ideal_kernel_change = self._compute_kernel_mapping(
self.connectivity_kernel,
control_template.kernel
)
return ideal_kernel_change
def find_optimal_stimulation_site(self, disease_discriminant):
site_effects = []
for site in range(self.num_nodes):
effect = self._simulate_stimulation_effect(site, disease_discriminant)
site_effects.append(effect)
optimal_site = np.argmax(np.(site_effects))
optimal_site, site_effects[optimal_site]
():
current_state = .reservoir.get_state(real_time_fc)
stimulation_amplitude = ._compute_optimal_amplitude(current_state)
stimulation_amplitude
Experimental Protocol
- Data Collection: Resting-state fMRI from AD patients and controls
- Model Fitting: Train subject-specific reservoir models on individual FC data
- Classification: Use functional read-out to classify AD vs controls
- Target Identification:
- Compute ideal distributed correction
- Test single-site stimulation at largest deviation site (fails)
- Identify optimal site by effect on disease discriminant (succeeds)
- Closed-loop Validation: Implement real-time controller and validate efficacy
Applications
Clinical Neuroscience
- Alzheimer's Disease Treatment: Personalized neuromodulation protocols
- Other Neurodegenerative Disorders: Extend to Parkinson's, Huntington's, etc.
- Neuropsychiatric Conditions: Apply to depression, schizophrenia with FC alterations
Computational Neuroscience
- Brain Network Modeling: Subject-specific network dynamics reconstruction
- Functional Connectivity Analysis: Lagged FC modeling beyond static correlations
- Therapeutic Target Discovery: Identifying network-responsive intervention sites
AI/ML Applications
- Personalized Medicine: Model-informed treatment optimization
- Reservoir Computing: Clinical applications of reservoir models
- Closed-loop Control: Real-time adaptive neuromodulation systems
Limitations and Considerations
Data Requirements
- Requires high-quality resting-state fMRI data
- Sufficient sample size for control template construction
- Individual variability may require large training datasets
Computational Complexity
- Subject-specific model fitting is computationally intensive
- Real-time closed-loop control requires efficient implementation
- Parameter tuning for optimal performance
Clinical Translation
- Requires validation in clinical trials
- Integration with existing neuromodulation devices
- Regulatory approval for personalized protocols
Activation Keywords
- Alzheimer's disease functional connectivity
- Reservoir computing neuromodulation
- Personalized brain stimulation
- Distributed connectivity signature
- Subject-specific network models
- Therapeutic responsiveness targeting
- Closed-loop neuromodulation
- Lagged functional connectivity
References
- Primary: Capone, C., Cece, E., Ciardiello, A., Gigante, G., Cisbani, E., & Mattia, M. (2026). From read-out geometry to in-silico stimulation: a distributed functional-connectivity signature of Alzheimer's disease. arXiv:2607.24356 [q-bio.NC].
- Related: Resting-state functional connectivity in neurodegenerative diseases
- Applications: Personalized neuromodulation and closed-loop brain stimulation
Verification Steps
- Collect resting-state fMRI data from AD patients and controls
- Implement subject-specific reservoir computing models
- Validate FC reconstruction accuracy
- Train AD classification read-outs
- Test single-site vs effect-based stimulation strategies
- Implement and validate real-time closed-loop controller
- Compare efficacy and dose requirements between approaches