MRI
ICAD 2010
Outline


ADNI 1 - Summarize most important results – internal
and external ADNI investigators
From ISAB
summary of reasoning and pilot work behind final
GO/ADNI 2 protocol
 Summarize QC procedures for new sequences
 Summarize parameters of sequences

Previous ADNI session



Dx, prediction, rates of change, sample size
Brewer/Dale – Freesurfer
Barnes/Fox
Schuff – all the above also APOE and CSF effects
ADNI 1 image analysis results





MRI has better longitudinal power to detect change than clinical
instruments, resulting in smaller sample sizes for clinical trials in
both MCI and AD patients
Provided sample size estimates for powering clinical trials for
MCI and AD, comparing various methods
Measurement method matters: some MRI analysis methods had
greater longitudinal power than others
best performing MRI measures overall  TBM, BSI, Freesurfer
greater white matter hyperintensity load in AD than control and
in MCI than control subjects who would be typically enrolled in
therapeutic trials [Carmichael]  use as co variate in clinical trials
Hua et al
Sample Sizes per Arm Needed to Power Treatment Study in
AD/MCI
Citation
Subjects
Source of subjects
Measurement
Method
Sample size required to detect
treatment effects
Holland et al 2009 (51)
129 AD
299 MCI
Referral sample ADNI
Ctx thickness ERC ROI
Assuming 24 month trial, 25%
effect size, 80% power, scans
every 6 mo; 45 AD per arm; 135
MCI per arm
Hua et al 2010 (52)
50 AD
122 MCI
Referral sample ADNI
TBM temporal lobe
Assuming 12 month trial, 25%
effect size, 80% power; 43 AD
per arm; 82 MCI per arm
Leung et al 2010 (53)
81 AD
Referral sample ADNI
KN-BSI
Assuming 12 month trial, 25%
effect size, 80% power; 81 AD
per arm
Schuff et al 2009 (54)
96 AD
226 MCI
Referral sample ADNI
Hippocampal volume (SNT),
model includes 3 scans,
Markov chain, APOE
Assuming 12 month trial, 25%
effect size, 90% power; 186 AD
per arm; 341 MCI per arm
Vemuri et al 2010 (50)
71 AD
149 MCI
Referral sample ADNI
Ventricular –BSI
Assuming 12 month trial, 25%
effect size, 80% power, 2-sided 2
sample t-test at 0.05; 100 AD per
arm; 186 MCI per arm
Wolz et al 2010 (55)
126 AD
279 MCI
Referral sample ADNI
Simultaneous 4D graph
Segmentation
Assuming 12 month trial, 25%
effect size, 80% power, 2-sided 2
sample t-test at 0.05; 67 AD per
arm; 206 MCI per arm
Cross Sectional Separation of Clinically Diagnosed AD vs. Controls
Citation
Subjects
Source of subjects
Measurement
Method
Results
Gerardin et al 2009 (27)
Neuroimage
CN 25
AD 23
Referral sample ADNI
Hippocampal shape metric
Sens 96%, spec 92%
Hinrichs et al 2009 (28)
CN 94
AD 89
Referral sample ADNI
Multi voxel classifier
AUROC 0.88
Kohannim et al 2010 (30)
CN 213
AD 158
Referral sample ADNI
Multi voxel classifier
AUROC 0.89
McEvoy et al 2009 (31)
CN 139 AD 84
Referral sample ADNI
Ct thickness; med and lat
temporal, isthmus cingulated
orbito frontal
Sens 83%, spec 93%
Walhovd et al. 2010 (32)
42 CN, 38 AD
Referral sample ADNI
Ct thickness
85%
ADNI 1 image analysis results



MRI rates of change in cognitively normal subjects are
greater in APOE e4 carriers than non-carriers [Schuff,
2009; Morra, 2009; Fjell, 2010]
lower CSF A42 was associated with a thinner cortex in
cognitive healthy controls [Tosun, 2010]
No difference between 1.5T and 3T in group-wise
discrimination or sample sizes needed to power trials
[Ho, 2009]
Association between low baseline CSF A1-42
concentrations and cortical thickness in cognitive
normal elderly
- Tosun, 2010
CSF AB and decreased brain volume in
cognitively normal elderly (CDR 0)
Fagan et al Annals 2009
Dynamic Biomarkers of the Alzheimer’s
Pathological Cascade
Jack et al, Lancet Neurol 2010; 9: 119-28
Ab Amyloid = CSF Ab42 or amyloid PET imaging; Tau Mediated Neuron Injury and Dysfunction =
CSF tau or FDG PET; Brain Structure = structural MRI
Annual change in global PIB ratio and ventricular volume by clinical diagnosis
PIB positive subjects (baseline global cortical PIB ≥ 1.5)
are represented with triangles and PIB negative subjects (baseline global cortical PIB < 1.5)
are represented with circles.
Jack et al, Brain 2009 132 (Pt 5):1355-65
ADNI: Predicting time to conversion from MCI to AD from baseline
biomarkers – univariately Vemuri et al Neurology 2009
Model summaries from age-adjusted Cox proportional hazards models of time from aMCI to AD. The
biomarker enters the model as restricted cubic spline with three knots.
Biomarker
Model χ2 (P)*
Nonlinearity
χ2 (P)†
Q3 vs. Q1
HR (95% CI) ‡
C-index§
STAND-score
19.0 (<0.001)
1.5 (0.22)
2.6 (1.7, 4.2)
0.69
Aβ1-42
8.2 (0.02)
5.4 (0.02)
0.8 (0.5, 1.3)
0.62
log(t-tau)
6.8 (0.03)
5.0 (0.03)
1.7 (1.1, 2.6)
0.60
log(p-tau181P)
6.6 (0.04)
1.5 (0.22)
1.8 (1.1, 2.9)
0.61
11.0 (0.004)
8.5 (0.004)
2.0 (1.1, 3.4)
0.62
log(t-tau/Aβ1-42)
* 3 degree of freedom likelihood ratio test of biomarker significance given that age is in the model
† 1 degree of freedom likelihood ratio test of nonlinearity of biomarker effect given that age is in the model
‡ Hazard ratio (95% CI) comparing the third quartile (75th percentile) to the first quartile (25th percentile) of the biomarker.
§The concordance index (C-index) in a survival model is analogous to the AUROC in a logistic model and represents the estimated probability the model will
correctly predict which of two patients has the longer time until conversion from aMCI to dementia.
Effect of APOE on
biomarkers
•AB chaperone
•Hypothesis is e4
would selectively
affect amyloid
biomarker
Vemuri et al, ePub Annals of
Neurology, 2010
Conclusions concerning image corrections
Metric is sample size per arm to detect a 25% rate reduction in
AD – question is, do image corrections reduce technical
variance in a meaningful way?



3D Grad warp ~ 10% SS reduction (Gunter, Med Phys,
2009)
Scaling correction ~ 12% SS reduction (Clarkson, NI,
2009), image registration (vs phantom) is preferred method
Intensity correction improves longitudinal precision – esp
multi array coils and 3T (Leow NI 2006; Boyes, NI 2008)
Conclusions ADNI phantom


value of scanner monitoring - 20% of all ADNI-1
scans would have been affected by errors of various
types had each scanner not been monitored
[Gunter, 2009]
Designed ADNI phantom. Has been adopted as the
starting point for quantitative MRI phantom by
ISMRM and NIST
Outline


ADNI 1 - Summarize most important results – internal
and external ADNI investigators
From ISAB
summary of reasoning and pilot work behind final
GO/ADNI 2 protocol
 Summarize QC procedures for new sequences
 Summarize parameters of sequences

GO/ADNI 2 MRI protocol - rational

ADNI 1 carry forward subjects: maintain MRI
methodological consistency



same 1.5T scanner, using ADNI 1 1.5T protocol
Discontinue dual 3T/1.5T scans for those in “3T arm”
new GO/ADNI 2 enrollees: modernize and expand MRI
protocol




3T
limit ~ 30-40 minutes (limits # possible sequences in protocol)
only product sequences – ie no WIPs
Core protocol on all scanners, and vendor specific “experimental” sub
studies
GO/ADNI 2 MRI 3T Protocol
1.
2.
3.
4.
5.
6.
3D T1 volume unaccelerated (MPRAGE Siemens and
Phillips, IR SPGR GE)
3D T1 volume 2X accelerated
FLAIR
long TE 2D gradient echo
Experimental: Siemens (ASL), GE (DTI), Phillips
(resting state EPI-BOLD)
Phantom (once per day if > 1 ADNI patient)
GO/ADNI 2 MRI Core Protocol


1.
1.
2.
All newly enrolled GO (and ADNI 2) subjects
All vendor systems
3D T1 volume (MPRAGE Siemens and Phillips, IR SPGR GE)
Each MRI exam will contain both an accelerated and a nonaccelerated 3D T1 acquisition – * not back-to-back, (3.6%) exam
“salvage rate”
FLAIR (instead of PD/T2) – better measures of WMH
long TE 2D gradient echo (e.g., TE = 20 ms) acquisition for
micro hemorrhage detection  what is natural history
(prevalence and incidence) if MCH and superficial siderosis in a
clinical trials population?
Phillips MPRAGE unaccel 9:06
Comparison
3T
Phillips MPRAGE accel 5:35
3D T1
35 yo
GE MPRAGE (ADNI-1) 9:17
volunteer
GE IR-FSPGR (ADNI-GO) 9:41
GE IR-FSPGR (ADNI-GO) 5:34
Long TE gradient echo scan
Micro
Hemorrhages
Superficial
Hemosiderosis
GO/ADNI 2 experimental sub-studies –
every subject gets one type






*vendor specific – each not done in every subject
Why? (1) impossible to standardize without WIPs, (2) no
product available, (3) limit 30 - 40 min
arterial spin labeling (ASL) perfusion - Siemens
diffusion tensor imaging (DTI) – GE
resting state functional connectivity – Phillips
purpose - to evaluate the feasibility of acquiring useful data
in a multi-center (but single vendor) setting - are these
techniques useful for clinical trials? Mission of ADNI.
QC of experimental sequences

Will only repeat exam for quality failure of un accelerated
3D T1
more scans with quality problems in GO/ADNI 2 than ADNI
1
 QC information will be more important


Raw images vs maps
Raw images – QCed at Mayo for protocol compliance,
completeness, head coverage, bulk motion, susceptibility
artifacts
 Maps QCed by individual labs (Thompson, DeCarli, Jack,
Schuff) that upload numeric data

DTI - Color Coded FA Maps
From Kantarci et al, in press Neurology
Page 1 of 10
GO/ADNI 2 protoocls
http://www.loni.ucla.edu/ADNI/Research/Cores/inde
x.shtml