HAI Book 2020 - Dec 18B

d with machine learning using
ical Alzheimer's disease

ng Jae Hwang2, Rebecca Koscik1, Tobey
on1, Barbara Bendlin1, Vikas Singh1

disease (AD) pathology may facilitate identifying
herapy. We propose a machine learning model which
1]PiB-PET and T1-weighted MRI.
he Wisconsin Registry for Alzheimer’s Prevention
least three time points of PiB-PET and T1-weighted
from 8 bilateral regions in the PET images known to
ssed using the Computational Anatomy Toolbox and
ectivity. We designed a differential equation/diffusion
ed connectivity networks and baseline PiB to predict

the model which characterizes amyloid propagation
vel connectivity matrix, in order to accurately predict
% of subjects were used for training from time-1 to
fold cross-validation was performed.
n the test set compared to group differences derived
etween two time points. Our model yields more
.6% of time-2 samples and 81.3% of time-3 samples.
s using only the baseline scans from each subject (MR
he linear estimate of the average group change.

76

7

77

7

78

Keywords: Alzheimer’s disease, Network diffusion, Di
connectivity

7

ifferential equations, PiB PET image, MRI
79

Heeman, Fiona

P17: [11C]PiB amyloid quantification

Fiona Heeman1, Janine Hendriks1, Bart N.M. va
Lammertsma1, Maqsood Yaqub1

1Amsterdam UMC, Vrije Universiteit Amsterdam, Rad
Neuroscience, Amsterdam, The Netherlands
Introduction: Reference tissue approaches for quanti
using [11C]PiB PET circumvent the need for arterial s
the patient and better suited for large clinical trials. Th
cerebellar grey matter (CBGM), may be compromised
possibly suffer from truncation or lower statistics. Thi
general full validation of these RRs against dynamic s
purpose of the present study was to evaluate the use o
using CBGM.
Methods: 43 subjects (17 AD, 13 MCI, 13 control) pa
sampling) or longitudinal studye were included. Dyna
were co-registered and time-activity curves (TACs) w
regularly used RR: CBGM, whole cerebellum, whole
subcortical white matter (ERWM). All RR TACs wer
Loganf and the simplified reference tissue modelg, and
variability, correlations and annual rates of change we
Results: Across reference tissue methods, whole cere
(maximum 2.77%, Table 1). Quantitative measures fo
those of the 2T4k_Vb derived DVR with CBGM (r>0
Furthermore, an inverted u-shape was consistently obs
RRs (Figure 1), similar to what has been observed bef
Conclusion: Across reference tissue approaches, the w
measuring amyloid using [11C]PiB PET, in particular

8

n and choice of reference region

an Berckel1, Isadora Lopes Alves1, Adriaan

diology and Nuclear Medicine, Amsterdam

ification of amyloid load in Alzheimer’s disease (AD)
sampling. These approaches are less burdensome to
he generally accepted reference region (RR),
d in late disease stages,a difficult to delineate and
is has resulted in the search for alternative RRs, but in
scans using the CBGM, is mostly lackingb,c. The
of alternative reference regions against dynamic scans

art of a single-centre test-retestd (six with arterial
amic [11C]PiB PET (90 minutes) and T1 MRI scans
were extracted for several regions, including five

brainstem, brainstem white matter/pons and eroded
re fitted using a plasma (2T4k_Vb) model, reference
d SUV ratios were calculated. Next, relative test-retest
ere calculated.
ebellum showed lowest test-retest variability
or all methods and RR showed good correlations with
0.80), but cerebellar RRs showed best slope.
served for rates of change computed with cerebellar
fore.h,i
whole cerebellum is the region of choice for
in a longitudinal setting.

80

Keywords: [11C]PiB, reference region, amyloid quan
8

ntification
81

Hwang, Seong Jae

P18: Retrospective prediction of amyl
risk-enriched Alzheimer’s disease coh

Seong Jae Hwang1, Rebecca Koscik2,3, Tobey B
Sterling Johnson2,3,5,6, Vikas Singh2

1University of Pittsburgh, Pittsburgh, PA, US
2University of Wisconsin-Madison, Madison, WI, US
3The Wisconsin Alzheimer’s Institute, Madison, WI, U
4University of Texas, Arlington, Arlington, TX, US
5Wisconsin Alzheimer’s Disease Research Center, Ma
6VA Geriatric Research, Education and Clinical Cent

Background: Understanding the longitudinal pattern
detection/intervention in Alzheimer’s disease (AD). T
crosses a critical threshold is one of the earliest signs
network to extrapolate the PiB-PET amyloid trajector
wise PAs, and investigate their associations with APO

Method: We measured the PiB-DVR in 8 AAL (left-r
participants from the Wisconsin Registry for Alzheim
longitudinal [C11]PiB-PET scans (mean=3.42/s.d.=1.
a sequential neural network called CRow which maps
predicts PiB-DVR trajectories beyond observed ages.
retrospectively predict their PAs. Then, we computed
global PiB-DVR threshold of 1.2 from prior studies. T
DVR threshold was considered as the PA (minimum=
subjects). For each region, we tested the difference of
ε4-allele) groups using the two-sample t-test.

Result: Fig. 1 shows the predicted trajectories (dotted
subjects in all 8 regions. Table 1 shows that all region
APOE groups (α=0.05 with Bonferroni correction). Fi
groups revealing less distinct PAs among the ROIs.

Conclusion: Early amyloid accumulation is one of the
our results support this by demonstrating its associatio
PiB-DVR trajectories and estimates the PAs in AD-re

8

loid accumulation trajectories in a
hort with sequential neural network

Betthauser2,5, Zirui Tao2, Won Hwa Kim4,

US
adison, WI, US
ter (GRECC), Madison, WI, US
of amyloid accumulation is crucial for early
The PiB+ Age (PA) when amyloid accumulation
of AD progression. We developed a sequential neural
ries retrospectively, estimate the subject- and region-
OE in a preclinical cohort.
right combined) regions of cognitively asymptomatic
mer’s Prevention (N=234) with at most four
.57 interval; mean=63.8/s.d.=6.7 age). We developed

PiB-DVR trajectories to corresponding ages and
We used all subjects to train CRow and
the region-wise PiB-DVR thresholds based on the
The age when PiB-DVR reaches the region-wise PiB-
=45, Wisconsin Life Expectancy Table for PiB-
f the PAs between APOE+ (ε4-allele) and APOE- (no

d) given the original trajectories (solid) of the PiB+
ns show significant differences in PAs between the
ig. 2 shows the boxplots of PAs between APOE

e promising avenues for preclinical AD detection, and
on with APOE. Our model accurately extrapolates the
elated regions.

82

8

83

Keywords: Preclinical AD, Amyloid Imaging, Longitu
8

udinal Analysis, Machine Learning
84

Jones, David

P19: Data-driven biological pattern sc
outperforms ROIs

David Jones1, Jonathan Graff-Radford1, Hugo B
Josephs1, Jennifer Whitwell1, Kejal Kantarci1, B
Knopman1, Val Lowe1, Clifford Jack1

1Mayo Clinic, Rochester, MN, US
Background: Hypotheses about the location of biolog
selection of ROIs based on Braak NFT staging. These
interest and are subject to off-target and partial volum
spatial information that can improve sensitivity. We p
biologically relevant tau-PET signal, that is robust to t
ROIs.

Methods: Using data from N=1750 scans from the M
normal aging and the spectrum of neurodegenerative d
FTD), we performed a data driven pattern analysis tec
variance Projection and Reduction (BPR). BPR biolog
entorhinal cortex (EC) and a temporal lobe meta-ROI

8

coring of Flortaucipir scans

Botha1, Matthew Senjem1, Heather Wiste1, Keith
Bradley Boeve1, Ronald Petersen1, David

gically relevant tau-PET signal have driven the
e ROIs may not capture the underlying biology of
me confounds. ROI methods do not leverage global
propose to use a data driven method of identifying
these confounds, and compare its performance to

Mayo Clinic MCSA, ADRC, and NRG studies of
diseases (e.g., AD, DLB, PCA, PPA, PSP, CBD, and
chnique we developed called Between-subject-
gical pattern-based scoring was then compared to
(MR) performance.

85

Results: BPR identified a global flortaucipir pattern t
interest in the cohort (Figure 1). Plots of flortaucipir s
high tau signal versus low signal using BPR (Figure 2
unimpaired (CU) participants that are PiB negative (C
strongly using BPR scoring and not at all with EC-bas
BPR based scoring showed superior AROC for discrim
CU/PiB- vs. CU/PiB+ (Figure 3).

8

that accounted for 93% of the biological signal of
signal versus age reveals superior group separation of
2 A, ERC; B, meta ROI; C, BPR). In cognitively
CU/PiB-), age is correlated with tau-PET signal most
sed scoring (Figure 2 D, ERC; E, meta-ROI; F, BPR).
minating PiB+ vs. PiB-, CU/PiB- vs. CI/PiB+, and

86

Conclusions: BPR is designed to be robust to off targ
capturing non-focal biological patterns of tau-PET sig
is superior to ROI scoring of tau-PET scans including
Keywords: Tau-PET, flortaucipir, ROI, Data, Pattern

8

get binding and partial volume confound while
gnal observed across the AD continuum. BPR scoring
g in amyloid negative CU participants.
n

87

Hahn, Alice

P20: Hippocampal volume mediates th
and amyloid-sensitive cognitive compo

Young Ju Kim1,2, Alice Hahn1,2, Soo Jong Kim1
Seo1,2,4,5,6

1Department of Neurology, Samsung Medical Center,
Korea
2Neuroscience Center, Samsung Medical Center, Seou
3Samsung Alzheimer Research Center, Samsung Medi
4Center for Clinical Epidemiology, Samsung Medical
5Department of Health Sciences and Technology, SAIH
6Clinical Research Design and Evaluation, SAIHST, S
7Departments of Neurology, Inje University College o

Background: Amyloidosis and the neurodegenerative
hippocampal volume (HV) were found to be highly as
essential to examine how these biomarkers are related
function in preclinical Alzheimer’s disease (AD). The
mediation effect of neurodegeneration exists between
normal (CN) elderly participants using the Preclinical

Methods: The MRI scans, PET scans, and the cogniti
Screening Battery-II of 373 CN elderly participants w
the z-scores of the Seoul Verbal Learning Test-delaye
the Korean Color Word Stroop Test-color reading, the
naming, and the Korean Mini-Mental State Examinati
principal component analysis. Then, we examined the
between amyloid positivity and the PASC.

Results: The results indicated that amyloid positivity
amyloid positivity was found to be associated with HV
Furthermore, bootstrapping analyses were used to sho
positivity and the PASC [95%CI (–0.301, -0.032), p<

Conclusions: In the present study, we found that HV
Thus, the comprehensive relationship of HV and amy
intervention of AD.

Keywords: preclinical Alzheimer’s disease, cognitive

8

he relationship between amyloidosis
osite in preclinical AD

1,2, Si Eun Kim7, Juhee Chin1,2,4, Sang Won

, Sungkyunkwan University School of Medicine, Seoul,
ul, Korea
ical Center, Seoul, Korea
Center, Seoul, Korea
HST, Sungkyunkwan University, Seoul, Korea
Sungkyunkwan University, Seoul, Korea
of Medicine, Haeundae Paik Hospital, Busan, Korea
e measures like cortical thickness (Cth) and
ssociated with cognitive function. Accordingly, it is
d to each other to impact on alterations of cognitive
erefore, the present study aimed to investigate if the
amyloidosis and cognitive function in cognitively
l Amyloid Sensitive Composite score (PASC).
ive test scores from the Seoul Neuropsychological
were used. First, we created the PASC by summating
ed recall, the Rey Complex Figure Test-delayed recall,
e Controlled Oral Word Association Test-animal
ion with different weight on each test derived by
e mediation effects of Cth and HV on the relationship

was directly associated with the PASC. Moreover,
V, which was further related to the PASC.
ow that HV mediates the relationship between amyloid
<.01].
mediated the effect of amyloidosis on the PASC.
yloidosis on cognitive function may be the key to early

composite, amyloid, neurodegeneration

88

Klein, Gregory

P21: Concordance of visual and quant
amyloid scans in the GRADUATE gan

Gregory Klein1, Paul Delmar2, Nicola Voyle3, J
Ovens4, Monika Baudler2, Paulo Fontoura2, Rac

1Roche Pharma Research and Early Development, Ba
2Roche/Genentech Product Development, Neuroscien
3Roche Products Ltd, Welwyn Garden City, UK
4InviCRO, LLC., Boston, MA, US

Background: Gantenerumab is a fully human monoc
Phase III trials (GRADUATE I/II [NCT03444870/ NC
Disease. Eligible participants must show confirmation
assessment of amyloid PET. This work compares visu

Methods: Florbetaben and flutemetamol were scanne
scan (4x5min) targeting 90min ± 1min post-injection.
flutemetamol were also allowed. PET scans were acqu
including NIA-AA criteria, MMSE ≥22, CDR-GS of

Visual reads were performed on the reconstructed, att
independent reads were performed, and a third read w

Quantitative results were obtained using Freesurfer (v
parietal, temporal and cingulate regions was used to c
cerebellar reference region. SUVR’s were translated i
24 was used for all three amyloid tracers.

Results: GRADUATE enrollment is still ongoing. PE
participants (744 florbetaben, 62 florbetapir, 163 flute
positive via visual assessment, versus 81.3% quantitat
11.8%. Rate of negative visual/positive centiloid was
0.4% (p < 0.001). Mean (SD) centiloid for the screene
visual-positive group was 5.1 centiloid units lower tha

Conclusions: Visual assessment of amyloid positivity
This should be considered when designing clinical tria

Keywords: amyloid, gantenerumab, positron emission

8

titative assessments of baseline
ntenerumab studies

Jacob Hesterman4, Ryan Petrulli4, Heather
chelle Doody2, Geoffrey A. Kerchner2

asel, Switzerland
nce, Basel, Switzerland

clonal antibody currently under evaluation in two
CT03443973]) for the treatment of early Alzheimer's
n of beta-amyloid pathology via CSF or visual
ual with quantitative screening results.
ed using 300 and 185 MBq respectively and a 20 min
. Historical scans of florbetaben, florbetapir, and
uired after the participant passed screening tests
0.5 or 1, and FCSRT inclusion criteria.
tenuation-corrected scans in PET subject space. Two
was employed in case of disagreement.
v6.0). A cortical volume-weighted sum of the frontal,
compute a standard uptake value ratio with a whole-
into the centiloid scale, and a positivity threshold of

ET screening scans to date were evaluated for 976
emetamol). Overall 70.3% of participants were
tively, resulting in an overall discordance rate of
11.4%, and for positive visual/negative centiloid was
ed population was 68.8 (42.2). Mean centiloid for the
an for the centiloid-positive group.
y is more conservative than quantitative assessment.
als using amyloid as an enrichment criterion.

n tomography

89

Kohli, Akshay

P22: Regional amyloid burden is assoc
discrete hippocampal sub-regions

Akshay Kohli1, Kao Lee Yang1, Nicholas M Vo
Andrew L. Alexander1,3, Bradley T Christian1, S
Barbara B Bendlin

1Wisconsin Alzheimer’s Disease Research Center, Un
Health, Madison, WI, US
2Geriatric Research Education and Clinical Center, W
Madison, WI, US
3Department of Medical Physics, University of Wiscon

Background: While neurodegeneration in Alzheimer
utilizing T1-weighted MRI are often subject to bias fie
limiting their reliability and reproducibility. In contras
metric of tissue relaxometry. MPnRAGE facilitates th
images for the precise estimation of T1. A previous stu
fluid Aß levels to widespread alterations in R1 (1/ T1).
(by [C-11]PiB-PET) on quantitative T1. We hypothesi
greater Aß burden, and that affected regions would dif
volume ratios (DVR).

Methods: 115 cognitively unimpaired participants (T
MRI (to derive T1). [C-11]PiB DVRs were calculated
were registered to a population template via ANTS, an
independent whole-brain voxelwise linear regressions
11]PiB DVR ROI and T1, controlling for age, sex, and
at p<0.001(uncorrected) with voxel extent>50.

Results: Higher Aß in each ROI was associated with
regional T1 effects were similar across [C-11]PiB ROI
orbitofrontal cortices was associated with higher T1 in
remaining—posterior—ROIs was associated with hig
These results did not survive FWE-correction.

Conclusion: While Aß accumulation appears to be br
posterior Aß differentially effect the hippocampus. Hi
including myelin, iron, various macromolecules, and i

9

ciated with higher quantitative T1 in

ogt1, Sanjay Asthana1,3, Tobey J Betthauser1,
Sterling C Johnson, Steven R Kecskemeti,

niversity of Wisconsin School of Medicine and Public
William S. Middleton Memorial Veterans Hospital,
nsin - Madison, Madison, WI, US
r’s disease is well studied, conventional measures
eld inhomogeneities and lower tissue contrast,
st, quantitative T1 mapping provides a standardized
he generation of hundreds of T1-weighted contrast
udy employing quantitative MRI linked cerebrospinal
. Here we investigate the effect of regional Aß burden
ized that T1 would be higher in individuals with
ffer spatially based on regional [C-11]PiB distribution

Table 1) underwent [C-11]PiB-PET and MPnRAGE
d in 8 bilateral regions of interest (Table 2). T1 maps
nd smoothed using a 2mm gaussian kernel. Nine
s were fit to examine relationships between each [C-
d APOE4 positivity. Statistical maps were thresholded

higher regional T1 (Table 2). While several of the
Is, Aß in the anterior cingulate and medial
n the right anterior hippocampus, whereas Aß in the
gher T1 in the right posterior hippocampus (Figure 1).

roadly associated with altered T1, anterior and
igher T1 signal may reflect changes to tissue content
inflammation.

90

9

91

Keywords: relaxometry, amyloid, quantitative T1, hip
9

ppocampus, MRI
92

Kothapalli, Satya

P23: Genetically informed quantitativ
reveals brain tissue in hippocampal su
Alzheimer's disease

Satya Kothapalli1, Tammie Benzinger1,2, Andre
Fagan2,3,4, Marcus Raichle1,3,4, John Morris2,3, D

1Department of Radiology, Washington University, St
2Knight Alzheimer’s Disease Research Center, Washi
3Department of Neurology, Washington University, St
4The Hope Center for Neurological Disorders, Washin

Rationale: Brain tissue atrophy (volume loss) serves a
However, postmortem histopathological studies show
significantly exceeds volumetric loss of tissue and tha
are lost. Hence, identifying neuronal loss in vivo befo
early diagnosis.

Methods: Here we use a recently developed genetical
(GI-qGRE) MRI (Wen, et al, PNAS-2018) for in vivo
hippocampal subfields. Forty-seven participants were
Research Center, classified into three groups: (1) norm
(2) preclinical AD (CDR=0, Amyloid=positive) (n=12
Amyloid=positive) (n=9).

Results: GI-qGRE allowed identification of two tissu
tissue with relatively preserved concentration of neuro
detected a greater loss of neurons as compared with at
neuronal loss ranged between 35%-55% while volume
exhibit asymmetry in hippocampal subfields - lower v
hippocampal subfields compared with left in all group
correlated with cognitive performance on a battery of
strongest correlation occurred in the right hippocampu
p<0.05 for atrophy). Data also showed a positive asso
fraction in right hippocampus (r=0.7; p<0.0001) as co
atrophy (r=-0.42. p<0.05).

Conclusion: GI-qGRE identifies cognitively-associat
measured by atrophy.

Keywords: Quantitative-Gradient-Recalled-Echo met
Hippocampal subfields, Atrophy

9

ve Gradient Recalled Echo MRI
ubfields void of neurons in mild

ew Aschenbrenner2,3, Manu Goyal1,3, Anne M
Dmitriy Yablonskiy1,2,4

t. Louis, MO, US
ington University, St. Louis, MO, US
t. Louis, MO, US
ngton University, St. Louis, MO, US

as an in vivo MRI biomarker of neuronal damage.
w that neuronal loss in Alzheimer’s disease (AD)
at memory decline in AD does not start until neurons
ore it can be measured by atrophy can be valuable for

lly-informed quantitative-Gradient-Recalled-Echo
o evaluation of neuronal damage caused by AD in the

recruited from the Knight Alzheimer Disease
mal (CDR=0 and Amyloid status=negative) (n=26);
2); and (3) very mild/mild AD dementia (CDR 0.5/1,

ue types: tissue void of neurons (“dark matter”) and
ons (“viable volume”). In AD dementia, GI-qGRE
trophy in all hippocampal subfields. In right subfields
e loss ranged between 15%-20%. Also, our results
viable volume, and higher dark matter fraction in right
ps. Consistently, viable volume was strongly
psychometric tests rather than the total volume. The
us (r=0.56, p<0.0001 for neuron loss vs. r=0.36,
ociation between brain tauopathy and dark matter
ompared with association between brain tauopathy and

ted neuronal loss in mild AD dementia greater than

thod, Alzheimer’s disease, Neuronal loss,

93

Landau, Susan

P24: Validation of highly sensitive and
thresholds using ADNI participants an

Susan Landau1, Deniz Korman1, Santiago Bullic
Koeppe1,3, William Jagust1

1University of California, Berkeley, Berkeley, CA, US
2Life Molecular Imaging, Boston, MA, US
3University of Michigan, Ann Arbor, MI, US

Objectives: Existing [18F] florbetaben-PET positivity
using a uniform PET processing pipeline. We used sev
compare them across samples, and validate them with

Methods: We carried out FreeSurfer-based quantifica
cerebellum normalization across several florbetaben s
study rated as visually positive or negative (Barthel et
obtained in end-of-life patients and later evaluated wit
2015), (3) 189 ADNI3 participants, and (4) 62 young
SocNuclMed 2015). We calculated thresholds in thes
them by examining agreement across methods, associ
centiloids.

Results: An ROC analysis of the autopsy sample resu
normalization) or 39.2 centiloids (Fig1A), consistent w
II and III samples and linearly transformed to ADNI p
mean+2SD in 75 visually-negative ADNI scans, calcu
Gaussian Mixture Modelling of the ADNI florbetaben
(whole cerebellum normalization) or 20.2 centiloids. C
more sensitive (1.08) threshold accounted for 14% gre
(flortaucipir) uptake among florbetaben+ individuals (
threshold (R2=.28) (Fig3B).

Conclusion: Using a uniform PET processing pipelin
strategizes in several samples and converged on sensit
between these thresholds is dependent on study-specif

9

d specific florbetaben positivity
nd young controls

ch2, Susan De Santi2, Andrew Stephens2, Robert

S

y thresholds have not been compared across samples
veral strategies to derive positivity thresholds,
h respect to tau PET in ADNI.
ation of cortical summary SUVRs using whole
samples: (1) 142 scans obtained in an FDA Phase II
t al. Lancet Neurol 2011), (2) 71 FDA Phase III scans
th CERAD criteria at autopsy (Sabri et al. Alz & Dem
healthy (27.7+/-5.1yrs) controls (Seibyl et al.
se samples using several strategies, and evaluated
iations with tau PET in ADNI, and conversion to

ulted in a threshold of 1.20 (whole cerebellum
with an ROC-based threshold calculated in the Phase
pipeline “units” (Fig1B). However, calculation of
ulation of mean+2SD in young controls (Fig2A), and
n sample (Fig2B) all converged on a threshold of 1.08
Categorization of ADNI florbetaben scans using the
eater variance (Fig3A) in entorhinal tau-PET
(R2=.42), compared with the more specific (1.20)

ne, we examined multiple threshold derivation
tive (1.08) and specific (1.20) thresholds. Selection
fic goals.

94

9

95

Keywords: amyloid positivity, florbetaben, centiloids,
9

, flortaucipir
96

Lao, Patrick

P25: Additive contribution of white m
neurodegeneration on cognitive declin
cohort of older adults

Patrick Lao1, Anthony Chesebro1, Juliet Colon1,
Manly1, Yaakov Stern1, Richard Mayeux1, Adam

1Columbia University, New York, NY, US

Introduction: The 2018 NIA-AA Alzheimer’s diseas
biomarker approach to explain AD development and p
framework is the flexibility to incorporate various bio
neurodegeneration profile. The objective of this study
matter hyperintensity (WMH) to amyloid and neurode
community-based cohort of older adults.

Methods: A subset of cognitively healthy participants
Non-Hispanic White/Non-Hispanic Black/Hispanic) f
Aging Project underwent baseline Florbetaben PET (a
T2-weighted FLAIR MRI (WMH volume), as well as
1.5 years (up to 6 visits). Mixed effects models tested
language, executing function, visuospatial ability) ove
amyloid and cortical thickness (model 2), and amyloid
adjusted for demographics (age, sex, education, race/e
likelihood information criteria (IC).

Results: Higher baseline amyloid SUVR was associat
visuospatial ability. Lower baseline cortical thickness
executive function, and visuospatial ability. Higher ba
with decline in executive function and visuospatial ab
directly associated with memory, it did improve the ov
by increasing the effect size of amyloid (Famyloid, model 2

Conclusion: The contribution of WMH to amyloid an
cognitive trajectories over time in memory, executing
inclusion of vascular biomarkers in models of cognitiv

Keywords: White matter hyperintensity, Amyloid, Cor
Community-based cohort

9

matter hyperintensity to amyloid and
ne in a diverse, community-based

, Kay Igwe1, Yian Gu1, Nicole Schupf1, Jennifer
m Brickman1

se (AD) research framework moves towards a multiple
progression more fully. An advantage of the research
omarkers into a full or partial amyloid-tau-
y was to determine the additive contribution of white
egeneration on cognitive decline in a diverse,

s (n=155; age=69-99yrs; 65% women, 30%/44%/26%
from the Washington Heights-Inwood Columbia
amyloid SUVR), T1-weighted (cortical thickness) and
s subsequent neuropsychological assessments every
changes of four cognitive domains (memory,
er time and their associations with amyloid (model 1),
d, cortical thickness, and WMH (model 3), all
ethnicity). Model fit was assessed with negative log

ted with faster decline in language, memory, and
s was additionally associated with baseline language,
aseline WMH volume was additionally associated
bility. Interestingly, while WMH volume was not
verall fit of the statistical model (ΔIC=19.7, p=5E-5)
2=18.4; Famyloid, model 3=30.8).
nd neurodegeneration improved the fit of longitudinal
g function, and visuospatial ability, supporting the
ve decline in AD.
rtical Thickness, Longitudinal cognitive decline,

97

Lao, Patrick

P26: White matter hyperintensities ar
stage regions

Patrick Lao1, Krystal Laing1, Kay Igwe1, Antho
Moreno2, Jose Luchsinger1, Adam Brickman1

1Columbia University, New York, NY, US
2State University of New York Downstate, New York, N
Introduction: There is mounting mechanistic evidenc
hyperphosphorylation and aggregation of tau. In huma
cerebrovascular disease, including white matter hyper
progression of mild cognitive impairment (MCI) and A
objective of this study was to examine the magnitude
tau burden, one of the neuropathological culprits of A
middle-aged adults.
Methods: Participants (n=106; age=65±3yrs; 67% wo
for quantification of WMH volume, as well as MK-62
standard uptake value ratio (SUVR; inferior cerebellar
injection) in an ongoing study. Total WMH volume w
SUVR was calculated on the voxel-level. Voxelwise a
pattern of MK-6240 SUVR associated with total WM
Results: Descriptively, a scatterplot between WMH v
that WMH are present in the absence of substantial tau
due to Wallerian-like degeneration related to AD. Vox
higher MK-6240 SUVR in frontal and parietal regions
comparisons, p<0.01).
Conclusions: In a community-based cohort of late m
burden in early-Braak stage regions, primarily associa
burden in late-Braak stage regions, primarily associate
evidence that WMH may have a direct role in AD dev

9

re related to tau burden in late-Braak

ony Chesebro1, William Kreisl1, Herman

NY, US
ce in mouse models that hypoperfusion leads to the
ans, image-based biomarkers of small vessel
rintensities (WMH), are associated with risk and
Alzheimer’s disease (AD)-type dementia. The
and spatial extent of associations between WMH and
AD, in a cohort of community-dwelling Hispanic late

omen; 100% Hispanic) underwent T2* FLAIR MRI
240 PET and T1 MRI for quantification of MK-6240
ar gray matter reference region; 90-110 min post-
was quantitated in cubic centimeters, while MK-6240
analysis (SPM12) was used to determine the spatial
MH volume.
volume and regional MK-6240 SUVR demonstrates
u burden, suggesting that WMH are not exclusively
xel-wise, higher WMH volume was associated with
s (no family-wise error correction for multiple

middle-aged Hispanics , there is evidence that tau
ated with aging, is not related to WMH, but tau
ed with AD, is related to WMH. These results provide
velopment and progression.

98

Keywords: tau PET, MK6240, White matter hyperinte
9

ensity, late middle-age, Hispanics
99

Laymon, Charles

P27: Probability template method for
PET

Charles Laymon1, Davneet Minhas1, Jeffrey Jam
Klunk1, Dana Tudorascu1, Sarah Royse1, Shahid

1University of Pittsburgh, Pittsburgh, PA, US
2University of Wisconsin, Madison, WI, US
3University of Cambridge, Cambridge, UK
Background: A component of the Alzheimer’s Biom
study of amyloid deposition in Down syndrome (DS)
analysis is to define regions-of-interest (ROIs) by parc
(FS). FreeSurfer, however, may not be ideal for DS su
atlas priors and who are prone to motion during MR a
Template (PT) [Svarer et al, 2005] approach for impro
Method: MR and [11C]PiB PET scans were acquired
first processed through the standard FS (version 5.3) p
scale, unusable-to-good). PT analysis began with sele
MR scans with good FS-based parcellations. Subject-
stripping the subject MR scan and (2) warping each te
result was 12 versions of every FS region warped to th
specific PT-ROI was generated as the region of maxim
volume equal to the average volume of the post-warp
rated and compared to FS-ratings. For subjects with a
sampled using ROIs from both methods and results w
Results: Twenty-one subjects had FS-based parcellati
PT parcellations. Correlation of PiB SUVR between t
Conclusion: While the PT-method cannot capture the
provide parcellations for cases in which FreeSurfer fa
FreeSurfer in the DS population.

10

analysis of Down Syndrome amyloid

mes1, Bradley Christian2, Ann Cohen1, William
d Zaman3, Benjamin Handen1

markers Consortium-Down Syndrome (ABC-DS) is the
using PET imaging. A standard method of PET
cellating the subject’s T1-MR scan using FreeSurfer
ubjects who exhibit variations in anatomy from FS-
acquisition. In this work, we evaluated a Probability
oving parcellation results.
by the ABC-DS in 196 DS subjects. MR scans were
pipeline, and parcellation results were rated (4-point
ection of twelve templates, skull-stripped (SPM12)
-specific PT-processing consisted of (1) skull-
emplate to the skull-stripped image (SPM8). The
he subject MR scan. For each FS region, the subject-
mum overlap of corresponding template regions with
template regions. PT-ROIs for all subjects were
acceptable FS- and PT-ratings, [11C]PiB PET was

were correlated.
ions rated as unusable. Of these, 13 had acceptable
the two methods was high (Figure 1).
e cortical detail of FreeSurfer, it is frequently able to
ails. The PT approach provides a useful alternative to

00

Keywords: PET, Image Analysis, MRI
10