A signal subspace approach for modeling the hemodynamic response function in fMRI.

Area of Science:

• Neuroimaging
• Functional Magnetic Resonance Imaging (fMRI)
• Biomedical Signal Processing

Background:

• fMRI analysis commonly employs models for the hemodynamic response function (HRF).
• Traditional HRF models (e.g., Gaussian, Gamma) require analysts to pre-select parameters.
• This a priori selection can limit the flexibility and accuracy of HRF characterization.

Purpose of the Study:

• To present a novel method for characterizing the HRF across a broad parameter range.
• To develop a flexible framework for HRF modeling and brain activation detection in fMRI.
• To improve detection sensitivity and reduce reliance on pre-defined HRF parameters.

Main Methods:

• Utilized principal component analysis (PCA) to derive three basis signals representing HRF variability.
• Developed signal subspaces based on these PCA-derived basis signals and stimulation patterns.
• Applied linear and nonlinear modeling for HRF identification and activation detection.

Main Results:

• The proposed signal subspace method demonstrated increased detection sensitivity compared to trigonometric subspaces in simulated fMRI data.
• Application to event-related and block design fMRI data yielded results consistent with previous studies.
• The method successfully detected brain activation without requiring a priori assumptions on HRF shape parameters.

Conclusions:

• The PCA-based basis signal approach offers a robust and flexible method for HRF modeling in fMRI.
• This technique enhances the sensitivity of brain activation detection.
• The approach is applicable to various fMRI experimental designs and modeling strategies, improving HRF identification.