hst 583 fmri data analysis and acquisition
DESCRIPTION
HST 583 fMRI DATA ANALYSIS AND ACQUISITION. A Review of Statistics for fMRI Data Analysis Emery N. Brown Massachusetts General Hospital Harvard Medical School/MIT Division of Health, Sciences and Technology December 2, 2002. Outline. What Makes Up an fMRI Signal? - PowerPoint PPT PresentationTRANSCRIPT
HST 583 fMRI DATA ANALYSIS AND ACQUISITION
A Review of Statistics for
fMRI Data Analysis
Emery N. Brown
Massachusetts General Hospital
Harvard Medical School/MIT Division of Health, Sciences and Technology
December 2, 2002
Outline
• What Makes Up an fMRI Signal?
• Statistical Modeling of an fMRI Signal
• Maxmimum Likelihoood Estimation for fMRI
• Data Analysis
• Conclusions
THE STATISTICAL PARADIGM (Box, Tukey)Question
Preliminary Data (Exploration Data Analysis)
Models
Experiment (Confirmatory Analysis)
Model Fit
Goodness-of-Fit not satisfactory
Assessment SatisfactoryMake an Inference
Make a Decision
Case 3: fMRI Data Analysis
Question: Can we construct an accurate statistical model to describe the spatial temporal patterns of activation in fMRI images from visual and motor cortices during combined motor and visual tasks? (Purdon et al., 2001; Solo et al., 2001)
A STIMULUS-RESPONSE EXPERIMENT
Acknowledgements: Chris Long and Brenda Marshall
What Makes Up An fMRI Signal?Hemodynamic Response/MR Physics i) stimulus paradigm
a) event-relatedb) block
ii) blood flow iii) blood volume iv) hemoglobin and deoxy hemoglobin contentNoise Stochastic i) physiologic ii) scanner noiseSystematic i) motion artifact ii) drift iii) [distortion] iv) [registration], [susceptibility]
Physiologic Response Model: Block Design
Gamma Hemodynamic Response Model
Physiologic Model:
Event-Related Design
0 20 40 60 80 100 1200
0.5
1
Flow Term
0 20 40 60 80 100 1200
0.5
1
Volume Term
0 20 40 60 80 100 1200
0.5
1
Interaction Term
0 20 40 60 80 100 120-0.2
0
0.2
0.4
0.6
Modeled BOLD Signal
fa=1 fb=-0.5
fc=0.2
Physiologic Model: Flow, Volume and Interaction Terms
Scanner and Physiologic Noise Models
DATA:
, …,1 Ty y
The sequence of image intensity measurements on a singlepixel.
fMRI Signal and Noise Model
= ( ) + vt ty h t
Measurement on a single pixel at time
t
t y
Physiologic response( )h t
Activation coefficient
Physiologic and Scanner Noise
vt for = , …,t 1 T
We assume the vt are independent, identically distributed
Gaussian random variables.
fMRI Signal Model
Physiologic Response
( ) = ( ) ( - )h t g u c t u du
( )g t hemodynamic response
( )c t input stimulus
Gamma model of the hemodynamic response
-( ) = 1 - tg t t e
Assume we know the parameters of g(t).
MAXIMUM LIKELIHOOD
Define the likelihood function ( ) = ( )L | y f y | , the joint
probability density viewed as a function of the parameter
with the data y fixed. The maximum likelihood estimate
of is ˆML
ˆ ( ) = arg max ( ) = arg max ( ).
ML y L | y logL | y
That is, ˆ ( )ML y is a parameter value for which ( )L | y
attains a maximum as a function of
for fixed y.
ESTIMATION
Joint DistributionT
2 2T t tt 12 2
y h1 1f y
22
=( - )
( | ) = exp -
Log Likelihood
=log ( | ) = log( ) - ( - )
= ( )
2 2 2Tt tt 1
2
T 1f y 2 y h /
2 2
Maximum Likelihood
ˆ
ˆˆ
-
= =
-ε =
=
= ( - )
12T Tt t tt 1 t 1
2 1 2Tt tt 1
h h y
T y h
GOODNESS-OF-FIT/MODEL SELECTION
An essential step, if not the most essential step in a data analysis,is to measures how well the model describes the data. This should be assessed before the model is used to make inferencesabout that data. Akaike’s Information Criterion
ˆML2 f | 2p- log (y ) +
For maximum likelihood estimates it measures the trade-off between maximizing the likelihood (minimizing
ˆML2 f |- log (y )
and the numbers of parameters p, the model requires.)
GOODNESS-OF-FIT
• Residual Plots:
ˆˆ = -t t ty h
• KS Plots:
ˆ ( )2t Ν 0,
We can check the Gaussian assumption with our K-S plots.
Measure correlation in the residuals to assess independence.
EVALUATION OF ESTIMATORS
Given w( ),y an estimator of based on = ( , …, )1 ny y y
Mean-Squared Error: [ ( ) - ] = Variance + bias2 2E w y
Bias= [ ( )]- ;unbiasedness [ ( )] =E w y E w y
Consistency: ( ) as (sample size) w y n
Efficiency: Achieves a minimum variance (Cramer-Rao Lower Bound)
FACTOIDS ABOUT MAXIMUM LIKELIHOOD ESTIMATES
•Generally biased.
•Consistent, hence asymptotically unbiased.
•Asymptotically efficient.
•Variance can be approximated by minus the inverse of the Fisher information matrix.
•If ̂ is the ML estimate of , then ˆ( )h is the ML
estimate of
( ).h
Cramer-Rao Lower Bound
2dE w y
dw y
f y-E
[ ( )]
Var[ ( )][ log ( | ]
CRLB gives the lowest bound on the variance of an estimate.
CONFIDENCE INTERVALS
The approximate probability density of the maximum
likelihood estimates is the Gaussian probability density withmean and variance -- ( ) 1I where ( )I is the Fisherinformation matrix
log (y )( ) = -
2
2
f |I E
An approximate confidence interval for a component of is
ˆ -± ( )
121
i,ML |z iiz I
THE INFORMATION MATRIX
-=
- -
( )( ) = -
(
2 1 2Ttt 1
2 2 1
h 0I
0 ) T 2
CONFIDENCE INTERVAL
ˆ ˆ
ˆ ˆ
-
=
-
±
±
12
12
2tTT 1
2 2
2 h
2 2 T
Kolmogorov-Smirnov Test White Noise Model
2 2
White Noise ModelPixelwise Confidence Intervals for the Slice
fMRI Signal and Noise Model 2
= ( ) + vt ty h t
Measurement on a single pixel at time
t
t y
Physiologic response( )h t
Activation coefficient
Physiologic and Scanner Noise
tv v t t -1 for = , …,t 1 T
We assume the vt are correlated noise AR(1)
Gaussian random variables.
Simple Convolution Plus Correlated Noise
Kolmogorov-Smirnov Test Correlated Noise Model
2 2
Correlated Noise ModelPixelwise Confidence Intervals for
the Slice
AIC Difference = AIC Colored Noise-AIC White Noise
fMRI Signal and Noise Model 3
t ty t= s( ) + v
Measurement on a single pixel at time
t
t y
Physiologic response
1cos( ) sin( )
q
rs t rt B rt
r r( ) = A
Physiologic and Scanner Noise
vt for = , …,t 1 T
We assume the vt are independent, identically distributed
Gaussian random variables.
Harmonic Regression Plus White Noise Model
AIC Difference Map= AIC Correlated Noise-AIC HarmonicRegression
Conclusions
• The white noise model gives a good description of the hemodynamic response
• The correlated noise model incorporates known physiologic and biophysical properties and hence yields a better fit
• The likelihood approach offers a unified way to formulate a model, compute confidence intervals, measure goodness of fit and most importantly make inferences.