A comparative evaluation of wavelet-based methods for hypothesis testing of brain activation maps.

Area of Science:

• Neuroimaging
• Statistical Signal Processing
• Biomedical Engineering

Background:

• Functional magnetic resonance imaging (fMRI) generates complex spatial-temporal data requiring sophisticated statistical analysis for activation mapping.
• Traditional methods may face challenges in controlling statistical errors while maintaining sensitivity in fMRI data analysis.
• Wavelet-based approaches offer multiresolution analysis capabilities, potentially improving signal detection and error control.

Purpose of the Study:

• To investigate the potential of wavelet-based hypothesis testing methods for activation mapping in human fMRI data.
• To evaluate the trade-off between Type I error control and statistical power (ROC curve analysis) for different wavelet methods.
• To compare the performance of wavelet thresholding/shrinkage methods against classical and Bayesian hypothesis testing frameworks.

Main Methods:

• Application of wavelet thresholding and shrinkage techniques for hypothesis testing on fMRI data.
• Implementation of false discovery rate (FDR) control for testing multiple wavelet coefficients from 2D discrete wavelet transform (DWT).
• Utilized change-point detection with recursive hypothesis testing and Bayesian methods incorporating signal sparseness models.
• Comparative evaluation using 'null' fMRI data (subject at rest) and experimental event-related finger movement task data.

Main Results:

• All three investigated wavelet-based algorithms demonstrated effective Type I error control.
• The False Discovery Rate (FDR) method exhibited the most conservative error control.
• The Bayesian wavelet method proved to be the most powerful, generating plausible brain activation maps.
• Established a generalized connection between multiresolution wavelet methods and traditional monoresolution Gaussian smoothing methods for fMRI analysis.

Conclusions:

• Wavelet-based methods are effective for hypothesis testing and activation mapping in fMRI data.
• These methods offer robust control over statistical errors and generate reliable brain activation maps.
• The Bayesian approach provides a powerful tool for fMRI analysis, balancing sensitivity and error control.