Show/Hide Menu
Hide/Show Apps
anonymousUser
Logout
Türkçe
Türkçe
Search
Search
Login
Login
OpenMETU
OpenMETU
About
About
Open Science Policy
Open Science Policy
Frequently Asked Questions
Frequently Asked Questions
Communities & Collections
Communities & Collections
Important issues for brain connectivity modelling by discrete dynamic bayesian networks.
Download
index.pdf
Date
2020
Author
Geduk, Salih
Metadata
Show full item record
This work is licensed under a
Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License
.
Item Usage Stats
13
views
11
downloads
To understand the underlying neural mechanisms in the brain, effective connectivity among brain regions is important. Discrete Dynamic Bayesian Networks (dDBN) have been proposed to model the brain’s effective connectivity, due to its nonlinear and probabilistic nature. In modeling brain connectivity using discrete dynamic Bayesian network (dDBN), we need to make sure that the model accurately reflects the internal brain structure in spite of limited neuroimaging data. Based on the fact that there are many dDBN structure learning applications in the recent literature and most of them use very limited amount of data, some facts should be made clear at least for the model convergence which depends on the number of data, the model complexity, and the learning approach. In this thesis, we analyzed the sample complexity of dDBN to find the required number of samples that guarantee successful learning. Firstly, we realized that the theoretical sample complexity for dDBN structure learning is not realistic, practical and applicable in practice. Therefore, we also focused on a practical and systematic approach for estimating the sample complexity for dDBN. Secondly, we evaluated the non-supervised discretization methods for functional magnetic resonance imaging (fMRI) data which has not been done yet to the best of our knowledge. We generated synthetic fMRI data that possess temporal relations. Then they were used for modeling effective connectivity by dDBN to compare the performance of each discretization method. Thirdly we analyzed the smoothing step of the fMRI data which is necessary to improve the signal to noise ratio. Experiments suggested that smoothing fMRI data with Gaussian function having a standard deviation to be 4 mm is suitable considering effective connectivity via dDBN. Lastly, by considering these results we used dDBN to model the brain connectivity of schizophrenia and control group. The results signify that schizophrenia is a disconnection syndrome.
Subject Keywords
Schizophrenia.
,
Keywords: Discrete Dynamic Bayesian Networks
,
fMRI
,
Structure Learning
,
Sample Complexity
,
Effective Connectivity
,
Schizophrenia.
URI
http://etd.lib.metu.edu.tr/upload/12625220/index.pdf
https://hdl.handle.net/11511/45454
Collections
Graduate School of Natural and Applied Sciences, Thesis