Optimizing preprocessing and analysis pipelines for single-subject fMRI: 2. Interactions with ICA, PCA, task contrast...

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Data Science

Background:

• Functional MRI (fMRI) preprocessing corrects artifacts like head motion and physiological noise.
• Preprocessing pipeline choices significantly impact fMRI results.
• The interaction between preprocessing and experimental design factors is not fully understood.

Purpose of the Study:

• To investigate how preprocessing choices interact with between-subject heterogeneity and task contrast strength in fMRI.
• To evaluate the impact of standard preprocessing and subspace estimation techniques (PCA, ICA) on fMRI data.
• To assess the performance of individual versus fixed preprocessing pipelines.

Main Methods:

• fMRI data from young, healthy adults performing an adapted Trail-Making Test with two cognitive contrast levels.
• Standard preprocessing steps including motion correction, noise reduction, and detrending.
• Subspace estimation using Principal Component Analysis (PCA) and Independent Component Analysis (ICA).
• Performance evaluation using Penalized Discriminant Analysis, reproducibility (R), and prediction (P) metrics.
• Simulation methods to assess bias from individual-subject optimization.

Main Results:

• Individual pipeline optimization showed no significant bias compared to fixed preprocessing.
• Task contrast significantly influenced fixed pipeline performance; PCA and ICA effects varied with contrast.
• Subject-specific pipeline optimization enhanced within-subject and between-subject overlap.
• Weaker cognitive contrasts were more sensitive to pipeline optimization.

Conclusions:

• fMRI result sensitivity depends on preprocessing choices and their interaction with experimental design factors.
• Individualized preprocessing pipelines offer advantages, particularly for weak signal contrasts.
• A quantitative denoising procedure is proposed, beneficial for small-sample and clinical fMRI datasets.