Evaluation and comparison of GLM- and CVA-based fMRI processing pipelines with Java-based fMRI processing pipeline...

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Data Analysis

Background:

• fMRI activation patterns depend heavily on preprocessing choices and statistical models.
• Previous evaluations of fMRI pipelines primarily used simulated data, with limited studies on real fMRI datasets.
• Understanding the impact of preprocessing on real fMRI data is crucial for reliable results.

Purpose of the Study:

• To investigate the effect of preprocessing options on GLM-based fMRI pipelines using real data.
• To optimize single-subject fMRI processing pipelines.
• To compare univariate GLM and multivariate CVA models for fMRI data analysis.

Main Methods:

• Utilized a novel fMRI processing pipeline evaluation system based on prediction accuracy and reproducibility.
• Investigated common fMRI preprocessing steps including slice timing correction, global intensity normalization, spatial smoothing, and temporal detrending.
• Compared univariate General Linear Model (GLM) with multivariate Canonical Variates Analysis (CVA) on block-design fMRI data.

Main Results:

• Spatial smoothing and temporal detrending significantly improved fMRI pipeline performance and are essential for robust analysis.
• Combined optimization of spatial smoothing and temporal detrending further enhanced pipeline performance.
• Multivariate CVA generally outperformed univariate GLM in prediction accuracy, while GLM demonstrated higher reproducibility.

Conclusions:

• Preprocessing steps like spatial smoothing and temporal detrending are critical for accurate fMRI activation detection.
• The choice between univariate GLM and multivariate CVA involves a trade-off between prediction accuracy and reproducibility.
• A consensus approach combining different models may yield more accurate fMRI activation patterns.