Thesis
Multiple current dipole estimation in a realistic head model using signal subspace methods
Washington State University
Master of Science (MS), Washington State University
2004
Handle:
https://hdl.handle.net/2376/232
Abstract
Neural activity in the human brain can be modeled as a volume conductor with current dipoles representing collections of neuronal sources. Determining the spatiotemporal characteristics of the sources from such models requires a solution to the inverse electrostatic problem. In this study, algorithms, R-MUSIC and RAP-MUSIC, based on a signal subspace method, were used to invert combinations of synchronous and asynchronous dipolar sources in an anatomically realistic head model. The source localization of the algorithms was analyzed at signal-to-noise ratios from 0 to 30 dB, for a set of rank-four source configurations. Both the algorithms have the same performance for all the configurations. Localization of independent sources was excellent, even at low signal-to-noise ratios, demonstrating the potential performance advantages of a spatiotemporal analysis over a purely spatial treatment. The algorithms use a correlation threshold below which it searches for synchronous sources. A fixed correlation threshold was found to be inadequate. A SNR dependent correlation threshold was estimated for a set of rank four configurations considered and performance of the algorithms was analyzed. Localization for configurations containing synchronous sources was substantially degraded at signal-to-noise ratios below 20 dB, demonstrating a need for improved methods to distinguish between asynchronous and synchronous sources. The performance was also observed for a pair of sources of equal power with correlation coefficient of 0, 0.2463, 0.5064, 0.7505, and 1.0 between them. The performance was very good above 10 dB for partially correlated sources. The algorithms are able to identify the independent portion of the time series can be used for source localization of partially correlated sources at higher SNRs. The computational complexity of multidimensional search for synchronous sources was successfully reduced by initially searching the pair of synchronous sources at a lower resolution cortical region. The solution was then found by directing the search to a locally constrained region at full resolution around the initial solution, without degrading the performance.
Metrics
2 File views/ downloads
11 Record Views
Details
- Title
- Multiple current dipole estimation in a realistic head model using signal subspace methods
- Creators
- Bhavana Katyal
- Contributors
- Paul H. Schimpf (Degree Supervisor)
- Awarding Institution
- Washington State University
- Academic Unit
- Electrical Engineering and Computer Science, School of
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University; Pullman, Wash. :
- Identifiers
- 99900525171501842
- Language
- English
- Resource Type
- Thesis