A least angle regression method for fMRI activation detection in phase-encoded experimental designs.

Area of Science:

• Neuroimaging
• Biostatistics
• Signal Processing

Background:

• General Linear Models (GLM) are standard for fMRI activation detection but require a predefined design matrix.
• GLM's fixed design matrix assumes uniform neural responses, which is often inaccurate for fMRI data.
• Limitations of GLM necessitate more flexible and adaptive approaches for robust fMRI analysis.

Purpose of the Study:

• To introduce a novel adaptive regression method for fMRI activation detection.
• To overcome the limitations of predefined design matrices in GLM-based fMRI analysis.
• To enhance the accuracy and flexibility of detecting brain activation in fMRI studies.

Main Methods:

• Utilized the least angle regression (LARS) method to adaptively select hemodynamic response models from a series of online-constructed models.
• Incorporated adaptive determination of slow drift terms in the design matrix based on fMRI responses for optimal fitting.
• Implemented the selected model using LARS combined with Moore-Penrose pseudoinverse (PINV) and fast orthogonal search (FOS) to account for drift effects.

Main Results:

• The proposed adaptive LARS method demonstrated superiority over the traditional GLM with a fixed design matrix for fMRI activation detection.
• The adaptive method showed improved performance in phased-encoded experimental designs.
• The new regression approach increased the degrees of freedom in the regression analysis, offering greater statistical power.

Conclusions:

• The developed adaptive regression method provides a novel and effective approach for fMRI activation detection.
• This method surpasses traditional GLM-based analyses in accuracy and flexibility.
• The adaptive model selection and drift term determination offer significant advantages for interpreting fMRI data.