Transcript Controlling for Scanner-to-Scanner Variance in Multicenter

Controlling for Scanner-toScanner Variance in Multicenter
FMRI Studies:
Using Signal-to-Fluctuation Noise
Ratio (SFNR) as a Covariate in
a Level 2 ANCOVA
Lee Friedman, Ph.D.
And
Gary Glover, Ph.D.
Short Version
• Longer version this afternoon at 3:30
Overall Goal
• Ideal goal is to literally merge the data
across sites/scanners
• To evaluate scanner/site effects on brain
activation patterns in multisite fMRI studies
• To determine the cause of these scanner
effects
• To develop methods to reduce such
scanner effects
Methods
• FBIRN – Phase I.
• 5 young healthy males traveled to 10 sites
• Underwent identical fMRI studies on 2
separate days
• Studied on multiple tasks – will emphasize
the Sensorimotor (SM) Task and the
Resting-State Scans
Block trials, 15s “on”, 15s “off”, 8 blocks.
On Block:
• Alternating contrast checkerboard.
• Binaural tones generated. This is an
auditorily annoying, mistonal scale.
• Subject performs bilateral finger
apposition at 3Hz, in time with visual and
auditory cues.
Off Block:
Fixation cross, silence, no motion.
4 runs, 4.25 min each
Sensorimotor
Task
Resting-State Scans
• Stare at fixation cross.
• Two runs, 85 TRs (4.25 min)
Resting State Scan
Scanner Differences on SFNR
Resting-State Data – Gray Matter
Low Field: F(scanner) = 103.2, p <.0001
High Field: F(scanner) = 15.6 , p <.0001
Relationship between SFNR-GM-REST and PVAF
Low Field : F = 28.8, df = 1,4, p = 0.0058
high field: F = 17.8, df = 1,4, p = 0.0135
PVAF is a measure
of activation effect
size.
It is the Model r2*100
With both the HRF
Regressor and its
Derivative in the
Model.
It is the Percent
Of Variance
Accounted For
(PVAF)
High Field
Low Field
Effect of SFNR-GM-REST Adjustment on
F-tests for the Scanner Effect
Low Field Scanners
Effect of SFNR-GM-REST Adjustment on
F-tests for the Scanner Effect
High Field Scanners
LOW FIELD SCANNERS:
ESTIMATED MEANS BEFORE AND AFTER SFNR ADJUSTMENT
13
12
NO ADJUSTMENT
CV = 0.40
ADJUSTMENT WITH SFNR-GM-REST
CV = 0.22
45% reduction
11
10
9
PVAF
8
7
6
5
4
3
2
1
0
UCSD D15T NMEX UCIR IOWA
UCSD D15T NMEX UCIR IOWA
HIGH FIELD SCANNERS:
ESTIMATED MEANS BEFORE AND AFTER SFNR ADJUSTMENT
30
NO ADJUSTMENT
CV = 0.25
ADJUSTMENT WITH SFNR-GM-REST
CV = 0.14
44% reduction
25
PVAF
20
15
10
5
0
BWHM MINN
MAGH
D40T
STAN
BWHM MINN
MAGH
D40T
STAN
Conclusion
• Using SFNR as a covariate markedly
reduces the effect of scanner on a
measure of activation effect size.
Longer Version
•Other Types of SFNR
Estimates not using the
Resting-State Scan data.
•More methodological detail
•More discussion of issues