Thresholding FMRI images
| Main Author: | |
|---|---|
| Language: | English |
| Published: |
The Ohio State University / OhioLINK
2004
|
| Subjects: | |
| Online Access: | http://rave.ohiolink.edu/etdc/view?acc_num=osu1097769474 |
| id |
ndltd-OhioLink-oai-etd.ohiolink.edu-osu1097769474 |
|---|---|
| record_format |
oai_dc |
| spelling |
ndltd-OhioLink-oai-etd.ohiolink.edu-osu10977694742021-08-03T05:49:14Z Thresholding FMRI images Pavlicova, Martina voxels test statistics null hypotheses BH procedure FDR Magnetic Resonance Imaging (MRI) has become an important technique of human brain mapping. The objective of human brain mapping is to identify areas in the subject's brain that are activated by some tasks. To perform human brain mapping, experimental data can be obtained using an MRI scanner, resulting in massive space-time functional MRI (FMRI) data. It is common to perform hypothesis tests to identify active areas of the brain. To assess whether the measured intensities of a voxel change according to the experimental task, a voxel is declared significant (active) if an associated test statistic exceeds some threshold. While the literature offers mainly voxel-wise testing methods, a fundamental problem remains: How to account for the large number of hypotheses by simultaneous thresholding without losing power or control over the rate of falsely activated voxels? In this dissertation, we focus on the problem of simultaneously thresholding large numbers of hypotheses. A relatively recent approach to thresholding is to control the false discovery rate (FDR), the proportion of voxels declared activate which are truly not activated. A simple thresholding procedure that controls the FDR was introduced by Benjamini and Hochberg (1995). Later, Shen at al. (2002)} proposed Enhanced FDR (EFDR) procedure which thresholds the discrete-wavelet transform of the test-statistic map with reduced number of hypotheses rather than the map itself. In this dissertation, we begin by sketching the physics underlying the MRI and then focus on improving the aforementioned thresholding methods. We develop a p-value adaptive thresholding (PAT) procedure which exhibits more suitable behavior. We establish basic theorems connecting the PAT and the BH procedures. We compare both procedures in simulation study using artificially-activated FMRI datasets. As a result, we combine the PAT procedure with EFDR procedure, developing the EFDR/PAT procedure. In a second simulation study, we show that the EFDR/PAT exhibits many improvements over other procedures while thresholding large numbers of test hypothesis. Finally, we investigate the performance of a two-sample test for comparing the voxel-wise response across two experimental conditions that does not rely on normality of the underlying data. The sensitivity and specificity of the tests are compared. 2004 English text The Ohio State University / OhioLINK http://rave.ohiolink.edu/etdc/view?acc_num=osu1097769474 http://rave.ohiolink.edu/etdc/view?acc_num=osu1097769474 unrestricted This thesis or dissertation is protected by copyright: all rights reserved. It may not be copied or redistributed beyond the terms of applicable copyright laws. |
| collection |
NDLTD |
| language |
English |
| sources |
NDLTD |
| topic |
voxels test statistics null hypotheses BH procedure FDR |
| spellingShingle |
voxels test statistics null hypotheses BH procedure FDR Pavlicova, Martina Thresholding FMRI images |
| author |
Pavlicova, Martina |
| author_facet |
Pavlicova, Martina |
| author_sort |
Pavlicova, Martina |
| title |
Thresholding FMRI images |
| title_short |
Thresholding FMRI images |
| title_full |
Thresholding FMRI images |
| title_fullStr |
Thresholding FMRI images |
| title_full_unstemmed |
Thresholding FMRI images |
| title_sort |
thresholding fmri images |
| publisher |
The Ohio State University / OhioLINK |
| publishDate |
2004 |
| url |
http://rave.ohiolink.edu/etdc/view?acc_num=osu1097769474 |
| work_keys_str_mv |
AT pavlicovamartina thresholdingfmriimages |
| _version_ |
1719426089047031808 |