Advanced Methods Chris Rorden Some slides from Peter Bandettini

Slides:



Advertisements
Similar presentations
fMRI Methods Lecture6 – Signal & Noise
Advertisements

When zero is not zero: The problem of ambiguous baseline conditions in fMRI Stark & Squire (2001) By Mike Toulis November 12, 2002.
1 st Level Analysis: design matrix, contrasts, GLM Clare Palmer & Misun Kim Methods for Dummies
Richard Wise FMRI Director +44(0)
Statistical Parametric Mapping
Perfusion-Based fMRI Thomas T. Liu Center for Functional MRI University of California San Diego May 19, 2007.
Introduction to Functional and Anatomical Brain MRI Research Dr. Henk Cremers Dr. Sarah Keedy 1.
1 Temporal Processing Chris Rorden Temporal Processing can reduce error in our model –Slice Time Correction –Temporal Autocorrelation –High and low pass.
1 Temporal Processing Chris Rorden Temporal Processing can reduce error in our model –Slice Time Correction –Temporal Autocorrelation –High and low pass.
Designing a behavioral experiment
Opportunity to Participate EEG studies of vision/hearing/decision making – takes about 2 hours Sign up at – Keep checking.
Principles of MRI. Some terms: –Nuclear Magnetic Resonance (NMR) quantum property of protons energy absorbed when precession frequency matches radio frequency.
More MR Fingerprinting
HST 583 fMRI DATA ANALYSIS AND ACQUISITION Neural Signal Processing for Functional Neuroimaging Emery N. Brown Neuroscience Statistics Research Laboratory.
Magnetic Resonance Imagining (MRI) Magnetic Fields
Realigning and Unwarping MfD
Efficiency in Experimental Design Catherine Jones MfD2004.
Principles of NMR Protons are like little magnets
fMRI introduction Michael Firbank
Opportunity to Participate
Opportunity to Participate EEG study of auditory attention – takes about 2 hours Sign up on sheet or
FMRI: Biological Basis and Experiment Design Lecture 15: CBF and Localization II CBF techniques Big veins and big voxels 1 light year = 5,913,000,000,000.
Volumetric Analysis of Brain Structures Using MR Imaging Lilach Shay, Shira Nehemia Bio-Medical Engineering Dr. Alon Friedman and Dr. Akiva Feintuch Department.
Psy 8960, Fall ‘06 Fieldmaps1 Fieldmaps and distortion What is a fieldmap? How can we predict distortion? How can we correct distortion?
Principles of MRI Some terms: – Nuclear Magnetic Resonance (NMR) quantum property of protons energy absorbed when precession frequency.
Arterial Spin Labeling at 7T - Double Edged Sword
Measuring Blood Oxygenation in the Brain. Functional Imaging Functional Imaging must provide a spatial depiction of some process that is at least indirectly.
Signal and Noise in fMRI fMRI Graduate Course October 15, 2003.
I NTRODUCTION The use of rapid event related designs is becoming more widespread in fMRI research. The most common method of modeling these events is by.
Magnetic Resonance Imagining (MRI) Magnetic Fields Protons in atomic nuclei spin on axes –Axes point in random directions across atoms In externally applied.
Efficiency – practical Get better fMRI results Dummy-in-chief Joel Winston Design matrix and.
GUIDE to The… D U M M I E S’ DCM Velia Cardin. Functional Specialization is a question of Where? Where in the brain is a certain cognitive/perceptual.
Statistical Parametric Mapping Lecture 9 - Chapter 11 Overview of fMRI analysis Textbook: Functional MRI an introduction to methods, Peter Jezzard, Paul.
An Example Using PET for Statistical Parametric Mapping An Example Using PET for Statistical Parametric Mapping Jack L. Lancaster, Ph.D. Presented to SPM.
Pulse Sequences Types of Pulse Sequences: Functional Techniques
Basics of Functional Magnetic Resonance Imaging. How MRI Works Put a person inside a big magnetic field Transmit radio waves into the person –These "energize"
Signal and noise. Tiny signals in lots of noise RestPressing hands Absolute difference % signal difference.
FINSIG'05 25/8/2005 1Eini Niskanen, Dept. of Applied Physics, University of Kuopio Principal Component Regression Approach for Functional Connectivity.
FMRI Methods Lecture7 – Review: analyses & statistics.
Perfusion MRI in GSK Study
Statistical Parametric Mapping
FMRI – Week 4 – Contrast Scott Huettel, Duke University MR Contrast FMRI Graduate Course (NBIO 381, PSY 362) Dr. Scott Huettel, Course Director.
fMRI Task Design Robert M. Roth, Ph.D.
Statistical Parametric Mapping Lecture 4 - Chapter 7 Spatial and temporal resolution of fMRI Textbook: Functional MRI an introduction to methods, Peter.
The General Linear Model (for dummies…) Carmen Tur and Ashwani Jha 2009.
Ch. 5 Bayesian Treatment of Neuroimaging Data Will Penny and Karl Friston Ch. 5 Bayesian Treatment of Neuroimaging Data Will Penny and Karl Friston 18.
Arterial spin labeling
Statistical Analysis An Introduction to MRI Physics and Analysis Michael Jay Schillaci, PhD Monday, April 7 th, 2007.
Statistical Parametric Mapping Lecture 2 - Chapter 8 Quantitative Measurements Using fMRI BOLD, CBF, CMRO 2 Textbook: Functional MRI an introduction to.
The linear systems model of fMRI: Strengths and Weaknesses Stephen Engel UCLA Dept. of Psychology.
FMRI and Behavioral Studies of Human Face Perception Ronnie Bryan Vision Lab
Accuracy, Reliability, and Validity of Freesurfer Measurements David H. Salat
The general linear model and Statistical Parametric Mapping II: GLM for fMRI Alexa Morcom and Stefan Kiebel, Rik Henson, Andrew Holmes & J-B Poline.
Advanced fMRI Methods John VanMeter, Ph.D. Center for Functional and Molecular Imaging Georgetown University Medical Center.
Robert W. McCarley, Presenter Cindy Wible, Marek Kubicki ( generated fMRI data), and Dean Salisbury (generated ERP data) Harvard, VA Boston Healthcare.
Neuroimaging with MRI: combining physics and physiology to understand brain function Rhodri Cusack.
Parameters which can be “optimized” Functional Contrast Image signal to noise Hemodynamic Specificity Image quality (warping, dropout) Speed Resolution.
BOLD functional MRI Magnetic properties of oxyhemoglobin and deoxyhemoglobin L. Pauling and C. Coryell, PNAS USA 22: (1936) BOLD effects in vivo.
Physiological correlates of the BOLD signal an Introduction.
Connecting Sound with the Mind’s Eye: Multisensory Interactions in Music Conductors W. David Hairston, Ph.D Advanced Neuroscience Imaging Research Lab.
functional magnetic resonance imaging (fMRI)
FMRI data acquisition.
Contrast and Inferences
FMRI experimental design and data processing
Slicer fMRI introduction
Zen and the art of nuisance variable maintenance
Signal and Noise in fMRI
MfD 04/12/18 Alice Accorroni – Elena Amoruso
The Cognitive Science Approach
Basics of fMRI and fMRI experiment design
Presentation transcript:

Advanced Methods Chris Rorden Some slides from Peter Bandettini Advanced fMRI designs Adaptation fMRI Sparse fMRI Resting State fMRI Advanced fMRI analysis ICA Effective and Functional Connectivity Analysis Alternative measures of activation Perfusion msMRI Comparing SPM to FSL Some slides from Peter Bandettini fim.nimh.nih.gov/presentations CABI talk 18 November 2009

Adaptation Designs (from Kanwisher) Show two stimuli in rapid succession. See if a brain region can discriminate if these stimuli are the same or different. Classically, regions show adaptation – less time to process same information twice in a row. a.ka. ‘repetition suppression’ paradigm.

Adaptation Designs FFA activates strongly to faces Does it discriminate – yes: we see adaptation response. Similar adaptation is not seen for chairs, so suggests special role in face processing.

Sparse fMRI Standard fMRI acquires data continuously. Loud noises can make it difficult to examine auditory stimuli. Sparse imaging includes a delay between each fMRI volume, so stimuli can be presented while scanner is silent. Continuous Time (sec) 10 Sparse Time (sec) 10

Sparse fMRI Typically, sparse design like a block design – each acquisition measures effect of single stimuli. Stimuli must be presented ~5sec prior to acquisition. Sparse designs have less power than continuous designs, and it is difficult to estimate latency of BOLD response. Due to T1 effects, Sparse designs can still have good power. BOLD Time (sec) 10

Resting State fMRI Resting state fMRI allows us to estimate natural connectivity between regions: which regions cycle together. Essentially, have individual lie in scanner resting while you collect a lot of fMRI data. Must covary out low frequency scanner drift as well as high frequency physiological noise.

Resting State Correlations Activation: hand movement Rest: seed voxel in motor cortex B. Biswal et al., MRM, 34:537 (1995)

Independent Component Analysis In conventional analysis, we see if a HRF predicts our behavioral design. FSL includes MELODIC for ICA, includes nice description: www.fmrib.ox.ac.uk/analysis/research/melodic/ In ICA, we decompose fMRI data into different spatial and temporal components. estimate the BOLD response. estimate artifacts in the data, then run conventional analysis on denoised data. find areas of ‘activation’ which respond in a non-standard way. analyse data for which no model of the BOLD response is available (e.g. resting state fMRI).

ICA vs Conventional Analysis Conventional analysis is confirmatory: does my model predict data. Results depend on model ICA is exploratory: Is there anything interesting in the data? Can give unexpected results. What is the potential of ICA? FSL includes melodic, so you can examine our data. Many use melodic to remove artifacts.

Classic fMRI detects all regions involved with task Connectivity Classic fMRI detects all regions involved with task Motor task would elicit motor cortex, cerebellum and supplementary motor area. It would be much more insightful if we could see the direction of connections Examples include Dynamic Causal Modelling

Psycho-physiological Interaction (from Henson) SPM{Z} Parametric, factorial design, in which one factor is psychological (eg attention) ...and other is physiological (viz. activity extracted from a brain region of interest) V1 activity time Attention attention V5 activity V1 no attention V5 Attentional modulation of V1 - V5 contribution V1 activity

Effective vs Functional Connectivity (Henson) No connection between B and C, yet B and C correlated because of common input from A, eg: A = V1 fMRI time-series B = 0.5 * A + e1 C = 0.3 * A + e2 Correlations: A B C A 1 B 0.49 1 C 0.30 0.12 1 Functional connectivity B 0.49 A Effective connectivity -0.02 2=0.5, ns. 0.31 C

SPM2 Dynamic Causal Modelling (Henson) Attention Effects Photic – dots vs fixation Motion – moving vs static Attenton – detect changes Photic .52 (98%) .37 (90%) SPC .42 (100%) .56 (99%) V1 .69 (100%) IFG .47 (100%) Büchel & Friston (1997) .82 (100%) Motion V5 .65 (100%) Attention modulates the backward-connections IFG→SPC and SPC→V5 The intrinsic connection V1→V5 is insignificant in the absence of motion Friston et al. (2003)

Functional Connectivity Observe which region’s activity correlates. Can be done while resting in scanner Hampson et al., Hum. Brain. Map., 2002

Perfusion imaging Use Gd or blood as contrast agent. Allows us to measure perfusion Static images can detect stenosis and aneurysms (MRA) Dynamic images can measure perfusion (PWI) Measure latency – acute latency appears to be strong predictor of functional deficits. Measure volume Can also measure task-related changes in blood flow (ASL), similar to fMRI.

ASL MR signal is based proportion of atoms aligned with the magnet. Slightly lower energy state aligned, so atoms preferentially align. More alignment in higher fields However, 180° pulse will reduce this signal. 3T Net Magnetization  = 3T NM after 180° pulse  =

Arterial Spin Labeling Tag inflowing arterial blood Acquire Tagged image Repeat scan without tag Acquire Control image Subtract Control image – Tagged image 2 1 The difference in magnetization between tagged and control images is proportional to regional cerebral blood flow http://www.umich.edu/~fmri/asl.html 4 3

Data from Trio Control We collect 16 slices 3.5x3.5x6mm Tagged Difference Mean of 73 differences We collect 16 slices 3.5x3.5x6mm TR 2.2sec (4.4sec for tag+control pair). TE=12ms (very little BOLD artifact). Not wise to collect ASL faster than 2sec (otherwise, not enough transit time between volumes. Wise to use slower TR for individuals with impaired perfusion (stroke).

TI and TR influence contrast time TI (Inversion Time) TR (Repeat Time) TR (Repeat Time) TI must be long enough for tagged blood to wash in to tagged slice TR must be long enough to allow tagged blood to wash out of control slice

TR Optimal TR depends on the individual’s blood transit time. ~2.4s, the ‘tagged’ image has more tagged blood than the control image. ~1.8s, very low contrast: tagged blood in both control and tagged image. ~1.2s reverse contrast: tagged blood does not reach slice until the control image (except fast arteries).

Blood Transit Time BTT varies in individuals If the TR is very short, the blood will not yet reach the capillary beds. Therefore, the control image can appear darker than the tagged image! In particular, very little signal when BTT matches TR. Transit time actually faster during active than rest. Either calculate BTT for each individual MRM, 57, 661-669 or use a long TR (4s, e.g. 8 s for control+tag pair)

Theory: Signal in ASL Tagged image: Inflowing inverted spins within the blood reducing tissue magnetization: more flow = darker Control: Inflowing blood has increased magnetization than saturated tissue: more flow = brighter Control Tagged Acquisition Perfusion Signal Control Tagged Observation Mumford et al. (2006)

BOLD and Perfusion ASL scans are designed to measure perfusion However, because they are T2* scans, they also have a BOLD artifact. To minimize BOLD, keep TE to a minimum BOLD is present in BOTH tagged and control image Because the tagged and control images are acquired several seconds apart, simple subtraction of tagged and control image is not a good idea for event related designs.

Analysis Strategies Simple subtraction Subtract tagged image from subsequent control image Halves the amount of samples (e.g. with 3sec TR, one sample every 6sec). Problem: leading edge and falling edge of HRF will have very different signal in control and tagged image: poor choice for event-related designs.

Analysis Strategies Inter-trial subtraction Subtract tagged image from control image acquired at the same interval after task onset. Halves the amount of samples (e.g. with 3sec TR, one sample every 6sec). Problem: events must be ordered to coincide with TRs (e.g. period of on-off blocks is an odd number of TRs).

Analysis Strategies FSL interpolates controlled and tagged images to estimate signal for both control and tagged images. The number of volumes is not halved,– analysis proceeds similar to fMRI data. Samples not completely independent, so DF is adjusted. The FSL difference signal is actually added to a mean image for all samples, so that the relative signal-noise is similar to fMRI

Analysis Easy to analyze ASL data with FSL: FSL is not optimal Select perfusion check box FSL simply subtract tagged image from neighboring control FSL is not optimal Control and tagged image are not acquired simultaneously Therefore, they sample different points of HRF. There are alternatives Sinc interpolate to estimate simultaneous signals (interp_asl) Intertrial subtraction: compare control image with tagged image that was collected at same delay after event (Yang et al, 2000). Add both tagged and control images in a single model (Mumford et al, 2006). In general, FSL approach only good for block designs.

Measuring the initial dip ‘Initial dip’ than signal increase seen 5 sec later. No venous artefacts Later overcompensation may not be specific (‘watering a garden for the sake of a thirsty flower’). Very small signal Difficult to realize benefit if you can’t achieve good spatial resolution. Remains controversial – best parameters unknown. 2 1 Time (seconds) 0 6 12 18 24

Higher spatial resolution Contrast to noise ratio dependent on volume of hydrogen: Standard T2* 3x3x3mm = 27mm3 1.5*1.5x2mm = 4.5mm3 = 17% of SNR However, for small structures or edges, higher resolution reduces partial volume effects. Therefore, higher resolution can improve % signal change observed For ideas on optimal voxelsize, see www.pubmed.com/17101280

Arterial Spin Labelling Benefits: Direct measure of blood flow Less drift: Better for assessment of very slow (>1min) changes. Data whiter (less dominated by low frequency noise) Signal more from tissue than veins. Less spatial distortion than BOLD (BOLD requires long TE without spin-echo) Perhaps better statistical power for group analysis (calibrated measure has less variability). Disadvantages Requires two images: tagged and subtraction, therefore TR is twice as long. Less statistical power for individual (fewer samples) Can not collect many slices: can only see portion of brain, normalization difficult (hurts group statistics)

Super high resolution Venous effects decrease with field strength (e.g. at 1.5T, capillary/venous ratio much smaller than at 7T). Higher SNR with 7T can allow very high resolution imaging: Example ocular dominance columns for left and right eye projection to visual cortex. 0.5x0.5x3mm (0.75mm3) www.pubmed.com/17702606 Spin-echo sequences (HSE T2) can be used as well as traditional GE T2* at these field strengths to detect BOLD.

Neural current MRI (Bandettini) In theory, MRI phase maps should show the direct neural firing as detected by MEG. Intracellular Current Magnetic Field Surface Field Distribution Across Spatial Scales

magnetic source/neural current MRI fMRI BOLD is very indirect measure. Can we directly measure brain activity? Neural firing influences magnetic field (e.g. MEG). Is this effect big enough to measure? Very controversial. Most designs do not remove BOLD confound Recent work not encouraging www.pubmed.com/19539040 Image Phasemap