Lecture07_DeconvolveModelHRF

Lecture07_DeconvolveModelHRF - This Time Modeling BOLD...

Info iconThis preview shows pages 1–3. Sign up to view the full content.

View Full Document Right Arrow Icon
1 This Time • Modeling BOLD – HRF…deconvolution – Derivative boost – Balloon model • Bayesian Approaches • Project Discussion Linear Deconvolution • With long inter-event intervals and no BOLD overlap, linear deconvolution is the same thing as event-related averaging. • With short inter-event intervals and BOLD overlap, deconvolution acts like averaging combined with “unmixing” of partially overlapping predictor curves (design matrix columns). Linear Deconvolution Neuronal activity is modeled as a sequence of point events. Linear Deconvolution Serences et al. 2004. Each column represents one specific time point in the haemodynamic time course for one event. Linear Deconvolution • Each column represents one time point in the time course of one event. • Beta weights are the deconvolved time course. Example Design Matrix Linear Deconvolution 1 2 3 4 5 6 1 2 3 4 5 6 Beta Weights / Time Courses
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
2 Linear Deconvolution • Jittering the inter- event interval makes the columns linearly independent. • Use this matrix in the GLM. Example Design Matrix Linear Deconvolution Pros: • Produces time course • Does not assume specific shape for haemodynamic function • Can use narrow jitter window (rec. exponential distribution) • Can separate correct vs. errors • Robust against sequencing bias (though not immune to it) • Compound trial types possible Cons: • Complicated • Unrealistic assumptions about maintenance activity • Seems to be sensitive to noise Linear Deconvolution Results Right Caudal Intraparietal Sulcus Left Supplementary Eye Field Cue Curves Response Curves Pro Anti Nogo Assumptions • BOLD signal is linear (superposition and scaling). • BOLD signal is stationary over time. • Noise is white, Gaussian, and stationary time. ALL of these assumptions are FALSE. Scaling Superposition Spatially varying delays in fMRI Spatially varying delays in fMRI • Scan Related (up to half the TR) Scan Related (up to half the TR) • Can be removed with timing correction Can be removed with timing correction • Brain related (up to 4 Brain related (up to 4 -5 seconds) 5 seconds) • We We ’re stuck with these re stuck with these • Ways to deal with Ways to deal with SVDs • Ignore them (will bias the model to regions which have the Ignore them (will bias the model to regions which have the predicted delay) predicted delay) • Use a more flexible model (basis function, FIR) Use a more flexible model (basis function, FIR) • The simplest of these is to add a derivative term to the HRF The simplest of these is to add a derivative term to the HRF • Allows the model to fit delays on the order of +/ Allows the model to fit delays on the order of +/ - 2 seconds 2 seconds • CAVEATE: this works for a fixed effects analysis (soaks up varia CAVEATE: this works for a fixed effects analysis (soaks up varia nce and nce and increases significance) but NOT for a random effects analysis (w increases significance) but NOT for a random effects analysis (w hich will hich will
Background image of page 2
Image of page 3
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 02/10/2010 for the course TBE 2300 taught by Professor Cudeback during the Spring '10 term at Webber.

Page1 / 16

Lecture07_DeconvolveModelHRF - This Time Modeling BOLD...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online