Both model-driven and data-driven approaches can be used to analyse fMRI data on group level but they have their respective advantages and disadvantages. A model-driven approach such as the GLM requires prior knowledge of the BOLD contrast while offering a simple way of performing inference. In contrast, ICA is a data-driven approach that instead takes full advantage of the data but in which it is harder to make meaningful inference. A method for combining GLM and ICA on group level utilising their respective strengths is therefore tested. The performance of the method highly relies on ICA’s ability to estimate accurate sources. Violation of ICA’s assumptions can potentially affect this ability and this aspect is investigated to get a broader understanding. The method is tested on both simulated and real fMRI data consisting of subjects with schizophrenia and healthy controls. The simulation shows high power but a higher false positive rate than expected. The method is able to find the brain regions that are typically active during a spatial working memory task, such as the posterior parietal cortex and the dorsolateral prefrontal cortex. It is further found that ICA performs especially poorly when the sources are uniformly distributed on the unit sphere.
Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:uu-531463 |
Date | January 2024 |
Creators | Sjösten, Lina |
Publisher | Uppsala universitet, Statistiska institutionen |
Source Sets | DiVA Archive at Upsalla University |
Language | English |
Detected Language | English |
Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
Format | application/pdf |
Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0015 seconds