Neural stochastic differential equation networks as uncertainty quantification method for EEG source localization | |
| Author | Wabina, Romen Samuel R. |
| Call Number | AIT Thesis no.DSAI-22-03 |
| Subject(s) | Uncertainty--Mathematical models Electroencephalography Stochastic models |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Data Science and Artificial Intelligence |
| Publisher | Asian Institute of Technology |
| Abstract | EEG Source Localization (ESL) remains a challenging problem given the uncertain con ductivity values of the volume conductor models (VCMs). As uncertain conductivities vary across individuals, these may significantly influence the EEG inverse solution, leading to higher localization errors and may cause clinical misdiagnosis of brain diseases. One possi ble approach to minimize localization errors is to quantify conductivity values using uncer tainty quantification (UQ) methods. The widely-known UQ methods involve Bayesian tech niques, such as Bayesian Active Learning (BAL) and Monte-Carlo Dropout (MCD). They use prior conductivity values to determine its posterior inference and estimate its best cali bration. However, using these techniques have its main drawbacks: (1) solving for posterior inference is intractable, and (2) choosing incorrect priors can lead to higher localization er rors. In this study, we used the Neural Stochastic Differential Equations Network (SDENet), a combination of dynamical system and deep learning techniques that primarily utilized the Wiener process to minimize conductivity uncertainties in the VCMs and improve the EEG inverse problem. We compared the SDENet against Bayesian UQ techniques for the EEG forward and inverse problems. Results revealed that SDENet and BAL could minimize lo calization errors against a non-quantified VCM. The SDENet was also able to generate the lowest localization errors compared to past works. Future research may incorporate new stochastic dynamical system-based deep learning techniques as a UQ method to address other uncertainties in the EEG Source Localization problem. |
| Year | 2022 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Data Science and Artificial Intelligence (DSAI) |
| Chairperson(s) | Chaklam Silpasuwanchai |
| Examination Committee(s) | Dailey, Matthew N.;Mongkol Ekpanyapong |
| Scholarship Donor(s) | AIT Scholarship |
| Degree | Thesis (M. Sc.) - Asian Institute of Technology, 2022 |