Predictive structural dynamic network analysis
| Autor: | Adrian Preda, Laura Ponto, Gregory Jicha, Ging-Yuek Hsiung |
|---|---|
| Rok vydání: | 2015 |
| Předmět: |
Male
Dynamic network analysis Models Neurological Machine learning computer.software_genre Article Neuroimaging Alzheimer Disease Image Processing Computer-Assisted Humans Computer Simulation Sensitivity (control systems) Baseline (configuration management) Dynamic Bayesian network Brain Mapping Network architecture business.industry General Neuroscience Brain Bayes Theorem Network dynamics Magnetic Resonance Imaging Nonlinear Dynamics Biomarker (medicine) Female Artificial intelligence business Psychology computer Follow-Up Studies |
| Zdroj: | Journal of Neuroscience Methods. 245:58-63 |
| ISSN: | 0165-0270 |
| DOI: | 10.1016/j.jneumeth.2015.02.011 |
| Popis: | Background Classifying individuals based on magnetic resonance data is an important task in neuroscience. Existing brain network-based methods to classify subjects analyze data from a cross-sectional study and these methods cannot classify subjects based on longitudinal data. We propose a network-based predictive modeling method to classify subjects based on longitudinal magnetic resonance data. New method Our method generates a dynamic Bayesian network model for each group which represents complex spatiotemporal interactions among brain regions, and then calculates a score representing that subject's deviation from expected network patterns. This network-derived score, along with other candidate predictors, are used to construct predictive models. Results We validated the proposed method based on simulated data and the Alzheimer's Disease Neuroimaging Initiative study. For the Alzheimer's Disease Neuroimaging Initiative study, we built a predictive model based on the baseline biomarker characterizing the baseline state and the network-based score which was constructed based on the state transition probability matrix. We found that this combined model achieved 0.86 accuracy, 0.85 sensitivity, and 0.87 specificity. Comparison with existing methods: For the Alzheimer's Disease Neuroimaging Initiative study, the model based on the baseline biomarkers achieved 0.77 accuracy. The accuracy of our model is significantly better than the model based on the baseline biomarkers ( p -value = 0.002). Conclusions We have presented a method to classify subjects based on structural dynamic network model based scores. This method is of great importance to distinguish subjects based on structural network dynamics and the understanding of the network architecture of brain processes and disorders. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect