Virtual Connectomic Datasets in Alzheimer's Disease and Aging Using Whole-Brain Network Dynamics Modelling

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Large neuroimaging datasets are crucial for clinical research, aiding in automated stratification, diagnosis, and prediction.
• Missing concurrent structural connectivity (SC) and functional connectivity (FC) measurements pose a significant challenge in analyzing these datasets.

Purpose of the Study:

• To address the issue of missing connectivity features in neuroimaging datasets.
• To introduce computational whole-brain network modeling strategies for virtual data completion.

Main Methods:

• Utilized computational whole-brain network modeling with linear and nonlinear simulations.
• Employed self-consistent simulations to infer "virtual FC" from empirical SC and "virtual SC" from empirical FC.
• Applied machine learning classification to assess the performance of models trained on virtual versus empirical data.

Main Results:

• Demonstrated the feasibility of virtual data completion for both "virtual FC" and "virtual SC" using ADNI and healthy aging datasets.
• Machine learning algorithms trained on virtual connectomes achieved discrimination performance comparable to those trained on empirical data.
• Algorithms trained on virtual connectomes successfully classified novel empirical connectomes.

Conclusions:

• Computational network modeling enables the generation of realistic virtual connectomic datasets to overcome missing data challenges.
• Virtual data completion strategies can significantly enhance the utility of neuroimaging datasets for clinical research and machine learning applications.
• The proposed methods allow for the creation of arbitrarily large virtual datasets with high-fidelity network connectivity information.