Full text
| |
| Author(s): |
Javier Oscar Giacomantone
Total Authors: 1
|
| Document type: | Master's Dissertation |
| Press: | São Paulo. |
| Institution: | Universidade de São Paulo (USP). Escola Politécnica (EP/BC) |
| Defense date: | 2005-10-07 |
| Examining board members: |
Hae Yong Kim;
Sergio Shiguemi Furuie;
Said Rahnamaye Rabbani
|
| Advisor: | Hae Yong Kim |
| Abstract | |
This dissertation presents the main statistical methods to analyse fMRI temporal series to detect activated regions and to characterise the error involved in this decision. Due to low signal to noise ratio, elaborate processing techniques are necessary to analyse functional images. Statistical techniques are usually applied on the temporal series obtained from fMRI, resulting in a Statistical Parametric Map (SPM), a 3-D image that makes it possible to determine the state of a voxel, as activated or not activated, and the statistical significance of the result. We proposed a new, simple an elegant method based on Robust Anisotropic Diffusion (RAD) to exploit the spatial correlation of activated regions of the human brain. The new method, named Robust Anisotropic Diffusion of Statistical Parametric Maps (RADSPM), allows one to obtain statistical maps that improve the determination of activated areas from noisy fMRI data. The new parametric statistical maps, based on the voxel spatial correlation of the fMRI image, reduce the classification error thus improving the significance of the results. We have tested the new technique in both simulated and real fMRI, obtaining sharp and noiseless SPMs with increased statistical significance. We compare results of the new RADSPM method with those of a classic method, the conventional correlation method. We calculate the involved errors and we present Receiver Operating Characteristics (ROC) curves for both methods, comparing the most important parameters for simulated fMRI data. We also evaluate the new method on real data of a visual fMRI experiment. (AU) | |