← Back to Course Home

Week 05: fMRI Experimental Design

Transforming cognitive hypotheses into rigorous fMRI experiments requires navigating severe statistical and physiological constraints. This week covers block versus event-related designs, General Linear Model (GLM) design matrices, HRF convolution, temporal filtering, and multiple comparison corrections.

Interactive Demos & Activities

• fMRI GLM & Convolution Studio

  Primary Design Studio · GLM Matrix, Collinearity & OLS

  Interactive paradigm design: compare Block Designs (high detection power) vs. Jittered Event-Related Designs (high estimation efficiency), visualize multicollinearity and Variance Inflation (VIF), and solve OLS regression with statistical contrast $t$-tests.

• MRI Artifact Simulator (Motion & Ghosting)

  Motion Corruptions · Phase-Encode Ghosting

  Explore the motion artifact mode to understand why head motion during phase-encoding corrupts k-space, producing severe ghosting and false activations in fMRI time series.

• Interactive BOLD Signal & HRF Simulator

  HRF Convolution & Event Timing

  Experiment with event spacing and observe how closely spaced stimuli result in non-linear BOLD summation, illustrating the trade-off between detection power and estimation efficiency.

• Flash-Lag Psychophysics Experiment

  Behavioral Paradigm Design · Sensory Presentation

  Experience a classic visual perception timing experiment that illustrates how behavioral psychophysics paradigms are parameterized and synchronized with scanner trigger pulses.

• 1D Waveform & Fourier Spectrum Synthesizer

  Web Audio Spectrum Engine · Temporal Filtering

  Explore frequency decompositions and harmonic content to understand why scanner drift (<0.01 Hz) and cardiac/respiratory noise are addressed using high-pass filters and physiological regressors.

Topics Covered

• Block Designs vs. Event-Related Designs: Maximizing detection power (epochs of sustained activity) versus estimation efficiency of individual trial shapes.
• The General Linear Model (GLM): Constructing design matrices ($Y = X\beta + \epsilon$), convolving stimulus boxcars with the double-gamma HRF, and testing contrast vectors.
• Preprocessing Pipeline: Slice timing correction, motion realignment, spatial normalization (MNI/Talairach space), and spatial smoothing.