Open Access Series of Imaging Studies (OASIS):

Cross-sectional MRI Data in Young, Middle Aged,

Nondemented, and Demented Older Adults

Daniel S. Marcus1, Tracy H. Wang1, Jamie Parker2, John G. Csernansky1,

John C. Morris1, and Randy L. Buckner2,3,4

Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021

Abstract
& The Open Access Series of Imaging Studies is a series of mag- included. Multiple within-session acquisitions provide extremely
netic resonance imaging data sets that is publicly available for high contrast-to-noise ratio, making the data amenable to a wide
study and analysis. The initial data set consists of a cross-sectional range of analytic approaches including automated computational
collection of 416 subjects aged 18 to 96 years. One hundred of analysis. Additionally, a reliability data set is included containing
the included subjects older than 60 years have been clinically 20 subjects without dementia imaged on a subsequent visit
diagnosed with very mild to moderate Alzheimer’s disease. The within 90 days of their initial session. Automated calculation of
subjects are all right-handed and include both men and women. whole-brain volume and estimated total intracranial volume are
For each subject, three or four individual T1-weighted magnetic presented to demonstrate use of the data for measuring differ-
resonance imaging scans obtained in single imaging sessions are ences associated with normal aging and Alzheimer’s disease. &

INTRODUCTION
data in an open-access fashion, the OASIS project can
The Open Access Series of Imaging Studies (OASIS) is provide such exemplar images to the neuroimaging com-
a project aimed at making magnetic resonance imaging munity. Finally, OASIS data can be used for educational
(MRI) data sets of the brain freely available to the purposes as examples or as working data sets for courses
scientific community. By compiling and freely distribut- and workshops.
ing MRI data sets, we hope to facilitate future discoveries A number of data-sharing procedures have been
in basic and clinical neuroscience. Specifically, the OASIS adopted by the OASIS project to ensure the integrity
project is intended to play a number of roles. First, and usability of the provided data.
OASIS images and related measures serve as data sets for
continued scientific exploration. Beginning with the ini- 1. Quality control. All images are carefully screened
tial set of images obtained from more than 400 individ- for acquisition and processing errors. Images with severe
uals with and without dementia across the adult life artifacts are excluded from the distribution. Images with
span, OASIS data sets are selected to encourage inves- visible artifacts typical of MR acquisitions are included.
tigation of high interest topics and provide data that This variability in quality is intended to provide tool
would be difficult for individual laboratories to acquire. builders with realistic target images.
Second, OASIS data are targets for researchers creating 2. Documentation. Detailed acquisition protocols and
and furthering analytic techniques. Because the images parameters, demographics, and supporting measures
are acquired from subjects over a range of ages and (e.g., Mini-Mental State Examination [MMSE] scores) are
health conditions, OASIS data can be used to test the provided for all included images.
robustness and validity of techniques across the full 3. Preparation. In addition to the raw acquisition data,
range of the human brain’s varied landscapes. Third, images representing common postprocessing procedures
OASIS data can be used as benchmark targets for com- (e.g., atlas registration, bias field correction) are distrib-
paring similar analytic techniques. Standard images pro- uted as a convenience and for comparison with similar
vide a common point of reference for demonstrating and postprocessing methods.
contrasting methods. By providing carefully screened 4. Anonymization. In compliance with privacy regula-
tions, distributed data are assigned random identification
numbers (IDs), and all identifying information are re-
1
Washington University, 2Harvard University, 3Massachusetts Gen- moved from the image files. All links between original and
eral Hospital, 4Howard Hughes Medical Institute random IDs are permanently destroyed before release.

D 2007 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 19:9, pp. 1498–1507
 5. Access. The data are available on DVD and acces- METHODS
sible for viewing and downloading via a dedicated Web Subjects
site. Third parties are encouraged to redistribute the
data in ways that add to its general availability. Subjects aged 18 to 96 years were selected from a larger
6. Ongoing support. The OASIS Web site will serve as database of individuals who had participated in MRI
a forum for announcements and discussions regarding studies at Washington University, based on the availabil-
the distributions and as a clearinghouse for third party ity of at least three acquired T1-weighted images, right-
derived data sets. hand dominance, and a recent clinical evaluation for
7. Unrestricted usage. OASIS data may be freely older adults. Young and middle-aged subjects were re-

Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021

used with no restrictions. Attributions to the OASIS pro- cruited from the Washington University community.
ject will lend to the credence of OASIS as a reference Older subjects with and without dementia were ob-
resource. Anyone wishing to distribute derived data will tained from the longitudinal pool of the Washington
be encouraged to include the original images. The University Alzheimer Disease Research Center (ADRC).
OASIS Data Use Agreement is available at www.oasis- The ADRC’s normal and cognitively impaired subjects
brains.org. were recruited primarily through media appeals and
word of mouth, with 80% of subjects initiating contact
In the remainder of this article, we describe the first with the center and the remainder being referred by
OASIS data set release. This release includes cross- physicians (Berg et al., 1982, 1998). All subjects partici-
sectional data from 416 individuals across the adult life pated in accordance with guidelines of the Washington
span age 18 to 96 years, including 100 individuals with University Human Studies Committee. Approval for pub-
very mild to moderate Alzheimer’s disease (AD) as lic sharing of the data was also specifically obtained.
diagnosed clinically and characterized using the Clini- All subjects were screened for inclusion in this release.
cal Dementia Rating (CDR) scale (Morris et al., 2001; Young and middle-aged adults were questioned before
Morris, 1993). Cross-sectional data have proven useful image acquisition by a trained technician about their
in studying normal and diseased aging for both global medical histories and use of psychoactive drugs. Older
(whole-brain) and local structure. In normal aging, adults, aged 60 and older, underwent the ADRC’s full
whole-brain volume decline begins in early adulthood clinical assessment as described below. Subjects with a
and accelerates in advanced aging (DeCarli et al., 2005; primary cause of dementia other than AD (e.g., vascular
Fotenos, Snyder, Girton, Morris, & Buckner, 2005; dementia, primary progressive aphasia), active neuro-
Sowell, Thompson, Leonard, et al., 2004; Durston, logical or psychiatric illness (e.g., major depression), se-
2003; Blatter et al., 1995). Preferential volume loss of rious head injury, history of clinically meaningful stroke,
gray matter (Raz, Williamson, Gunning-Dixon, Head, & and use of psychoactive drugs were excluded, as were
Acker, 2000) and regionally specific thinning of the cor- subjects with gross anatomical abnormalities evident in
tex are also noted (Salat et al., 2004). Level of educa- their MRI images (e.g., large lesions, tumors). However,
tion, sex, socioeconomic status, and cardiovascular subjects with age-typical brain changes (e.g., mild atro-
health have been identified as contributing factors phy, leukoaraiosis) were accepted. MRI acquisitions typi-
in volume decline in advanced aging, suggesting that cally were obtained within a year of a subject’s clinical
subclinical health conditions contribute to age-related assessment (mean = 105 days, range = 0–314 days).
changes in brain structure (Fotenos, Mintun, Snyder, Three subjects with AD were scanned after a some-
Morris, & Buckner, in press; Raz, Rodrigue, & Acker, what longer duration (635, 449, and 443 days). The
2003; Koga et al., 2002; Coffey, Saxton, Ratcliff, Bryan, final data set includes 416 subjects. Twenty subjects in
& Lucke, 1999; Coffey et al., 1998). Individuals with their twenties were imaged on a second occasion within
clinically diagnosed AD show substantially reduced over- 90 days of their initial session to provide a means for
all brain volumes relative to age-matched peers as assessing reliability of data acquisition and analysis.
well as regional volume loss that has been well docu- Portions of the clinical, demographic, and image data
mented in the hippocampal formation, among other obtained from subjects in this release have been used in
regions (e.g., Jack et al., 1997; Raz et al., 1997; Killiany previous studies (Buckner et al., 2004, 2005, in press;
et al., 1993; Jack, Petersen, O’Brien, & Tangalos, 1992; Burns et al., 2005; Fotenos et al., 2005, in press; Head,
Scheltens et al., 1992). Cross-sectional MRI data have Snyder, Girton, Morris, & Buckner, 2005; Salat et al.,
also been used to facilitate and validate analytic tools 2004) and made publicly available, using different anon-
and algorithms, including atlas-based head size normal- ymous identifiers, via the fMRI Data Center under acces-
ization (Buckner et al., 2004), automated whole-brain sion number 2-2004-1168x.
segmentation (Fischl et al., 2002), automated labeling
of cerebral cortex into regions of interest (Desikan
Clinical Assessment
et al., 2006), and large-deformation maps of brain struc-
ture shape (Csernansky, Wang, Joshi, Ratnanather, & Dementia status was established and staged using the
Miller, 2004). CDR scale (Morris et al., 2001; Morris, 1993). The deter-

Marcus et al. 1499

mination of AD or control status is based solely on clinical Table 1. MRI Acquisition Details
methods, without reference to psychometric perfor- Sequence MP-RAGE
mance, and any potential alternative causes of dementia
(known neurological, medical, or psychiatric disorders) TR (msec) 9.7
must be absent. The diagnosis of AD is based on clinical TE (msec) 4.0
information (derived primarily from a collateral source) Flip angle (8) 10
that the subject has experienced gradual onset and pro-
gression of decline in memory and other cognitive and TI (msec) 20
functional domains. Specifically, the CDR is a dementia- TD (msec) 200

Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021

staging instrument that rates subjects for impairment
Orientation Sagittal
in each of the six domains: memory, orientation, judg-
ment and problem solving, function in community affairs, Thickness, gap (mm) 1.25, 0
home and hobbies, and personal care. Based on the Slice number 128
collateral source and subject interview, a global CDR
score is derived from individual ratings in each domain. Resolution (pixels) 256  256 (1  1 mm)
A global CDR of 0 indicates no dementia, and CDRs of
0.5, 1, 2, and 3 represent very mild, mild, moderate, and
severe dementia, respectively. These methods allow for
the clinical diagnosis of AD in individuals with a CDR of ages (Mugler & Brookeman, 1990) were acquired on
0.5 or greater, based on standard criteria (McKhann et al., a 1.5-T Vision scanner (Siemens, Erlangen, Germany)
1984), that is confirmed by histopathological examination in a single imaging session. Head movement was mini-
in 93% of the individuals (Berg et al., 1998), even for those mized by cushioning and a thermoplastic face mask.
in the earliest symptomatic stage (CDR 0.5) of AD who Headphones were provided for communication. A vi-
elsewhere may be considered to represent ‘‘mild cogni- tamin E capsule was placed over the left forehead to
tive impairment’’ (Storandt, Grant, Miller, & Morris, 2006). provide a reference marker of anatomical side. Posi-
tioning was low in the head coil (toward the feet) to
optimize imaging of the cerebral cortex. MP-RAGE pa-
Image Acquisition
rameters were empirically optimized for gray–white
For each subject, 3–4 individual T1-weighted magne- contrast (Table 1). An example of typical images is illus-
tization prepared rapid gradient-echo (MP-RAGE) im- trated in Figure 1.

Figure 1. Typical MRI data set. (A) Individual scan before defacing. (B) Same scan after defacing. Note that the defacing process leaves the
cranial vault intact while identifying facial features are removed. (C) Averaged motion-corrected image. Note improved signal-to-noise ratio.
(D) Atlas-registered gain-field-corrected image. (E) Tissue classification image.

1500 Journal of Cognitive Neuroscience Volume 19, Number 9

Postprocessing previously (Fotenos et al., 2005; Buckner et al., 2004).
In brief, estimated total intracranial volume (eTIV) was
For each subject, the individual scan files were converted
computed by scaling the manually measured intracranial
from Siemens proprietary IMA format into 16-bit Analyze
volume of the atlas by the determinant of the affine
7.5 format using a custom conversion program. Sensitive
transform connecting each individual to the atlas. This
header fields (patient ID, experiment date) were left
method has been shown to be proportional to manually
blank. Identifying facial features were then removed from
measured total intracranial volume (TIV, r = 0.94) and
the images using software provided by the fMRI Data
minimally biased by atrophy (Buckner et al., 2004).
Center (www.fmridc.org). This method first masks out all
Normalized whole-brain volume (nWBV) was comput-
nonbrain voxels in the image, dilates and smoothes the

Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021

ed using the FAST program in the FSL software suite
mask in an iterative fashion, and finally applies the
(www.fmrib.ox.ac.uk/fsl) (Zhang, Brady, & Smith, 2001).
enlarged mask to the original image to safely remove
The image was first segmented to classify brain tissue as
voxels outside the cranial vault. In 12 cases where this
cerebral spinal fluid, gray matter, or white matter. The
method either left too much of the face present or
segmentation procedure iteratively assigned voxels to
invasively cut into the cranial vault, de-identification was
tissue classes based on maximum likelihood estimates of
done using software provided by the Morphometry Bio-
a hidden Markov random field model. The model used
medical Informatics Research Network (www.nbirn.net).
spatial proximity to constrain the probability with which
This method registers the image to an atlas in which facial
voxels of a given intensity are assigned to each tissue
features and brain tissue have been hand-labeled. All
class. Finally, nWBV was computed as the proportion
voxels in the image that have a zero probability of being
of all voxels within the brain mask classified as gray or
brain and a nonzero probability of being face are assigned
white matter. The unit of normalized volume is percent,
an intensity value of zero. The de-identified images are the
which represents the percentage of the total segmented
images included in the present distribution and were used
voxels within the eTIV (Fotenos et al., 2005).
in all subsequent processing and analyses described here.
The images were then corrected for interscan head
movement and spatially warped into the atlas space of Quality Control
Talairach and Tournoux (1988) using a rigid transforma-
All images in the set were carefully screened for artifacts,
tion that differs in process from the original piecewise
acquisition problems, and processing errors. During the
scaling. The resulting transformation nonetheless places
screening process, each image was viewed on a per-slice
the brains in the same coordinate system and bounding
basis along each principal axis. Each image was assigned
box as the original atlas. The template atlas used here
a value from 1 to 3, with 1 indicating a high-quality im-
consisted of a combined young-and-old target previously
age, 2 indicating a scan with minor flaws, and 3 indicat-
generated from a representative sample of young subjects
ing severe flaws. Typical flaws included electronic noise
(n = 12) and old subjects without dementia (n = 12).
resulting in bright lines through multiple slices, motion
The use of a combined template has been shown to mini-
artifacts appearing as hazy bands across the image, poor
mize the potential bias of an atlas normalization proce-
head positioning resulting in wraparound artifacts, dis-
dure to overexpand atrophied brains (Buckner et al.,
tortions from dental work, and limited image contrast
2004). For registration, a 12-parameter affine transforma-
(Figure 2). Images with a rating of 3 were excluded from
tion was computed to minimize the variance between the
the data set. A number of borderline images remain in
first MP-RAGE image and the atlas target (Snyder, 1996).
the distribution, providing tool builders and testers with
The remaining MP-RAGE images were registered to the
a realistic range of acquisition quality. In cases where
first (in-plane stretch allowed) and resampled via trans-
individual scans were deemed unusable, the single scan
form composition into a 1-mm isotropic image in atlas
was removed from the set but the remainder of the
space. The result was a single, high-contrast, averaged
subject’s scans was included. Overall, 10 individual scans
MP-RAGE image in atlas space. Subsequent steps included
and 15 complete sessions were excluded from the final
skull removal by application of a loose-fitting atlas mask
release due to poor image quality.
and correction for intensity inhomogeneity due to non-
uniformity in the magnetic field. Intensity variation was
corrected across contiguous regions, based on a quadratic
inhomogeneity model fitted to data from a phantom RESULTS
(Styner, Brechbuhler, Szekely, & Gerig, 2000). Overview of the Data Set
The current data set consists of subjects across the adult
life span (Table 2). It includes 218 subjects aged 18 to
Estimated Total Intracranial Volume
59 years and 198 subjects aged 60 to 96 years. The over-
and Normalized Whole-brain Volume
all age distribution is presented in Figure 3. Each group
The procedures used for measuring intracranial and includes an approximately equal number of male and
whole-brain volumes have been described in detail female subjects. Of the older subjects, 98 had a CDR

Marcus et al. 1501

Figure 2. Example images
illustrate the quality control
rating procedures. (A) A scan
that received a rating of 1,
indicating a high-quality image
with minimal artifacts and
high contrast. (B) A scan that
received a rating of 3 because
significant motion artifacts.
Such artifacts are particularly
apparent when scrolling

Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021

through multiple image
slices. (C) A scan that
received a rating of 2 because
of significant in-plane
wraparound. Because the
nose does not actually overlap
into brain tissue, this scan
was retained in the data set.
Such borderline images were
relatively rare. (D) A scan that
received a rating of 3 because
of severe in-plane wraparound
that cuts into the brain.

score of 0, indicating no dementia, and 100 had a CDR The slope of the best fit line was 1.00 (r = 1.00) for eTIV
score greater than zero (70 CDR = 0.5, 28 CDR = 1, 2 and was 1.00 (r = 1.00) for nWBV. The mean absolute
CDR = 2), indicating a diagnosis of very mild to mod- percent differences before and after defacing were
erate AD. Additional demographics and clinical charac- 0.024% for eTIV and 0.059% for nWBV. The defacing
teristics of the older subjects are shown in Table 3. process, therefore, has little impact on morphometric
measures of interest.
Having established that defacing has minimal impact,
Anatomical Characteristics
eTIV and nWBV are plotted across the adult life span in
As a means of determining whether defacing techniques Figure 4 for the data included as part of the OASIS dis-
impact morphometric analysis, eTIV and nWBV were tribution. Analyses are selective and used to illustrate prop-
calculated before and after defacing for each subject. erties of the data. More detailed normative estimates of

Table 2. Age and Diagnosis Characteristics of the Data Set

Without Dementia With Dementia

Age Group Total n n Mean Male Female n Mean Male Female CDR 0.5/1/2

<20 19 19 18.53 10 9 0 0 0 0/0/0

20s 119 119 22.82 51 68 0 0 0 0/0/0
30s 16 16 33.38 11 5 0 0 0 0/0/0
40s 31 31 45.58 10 21 0 0 0 0/0/0
50s 33 33 54.36 11 22 0 0 0 0/0/0
60s 40 25 64.88 7 18 15 66.13 6 9 12/3/0
70s 83 35 73.37 10 25 48 74.42 20 28 32/15/1
80s 62 30 84.07 8 22 32 82.88 13 19 22/9/1
90 13 8 91.00 1 7 5 92.00 2 3 4/1/0
Total 416 316 119 197 100 41 59

1502 Journal of Cognitive Neuroscience Volume 19, Number 9

 Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021
Figure 3. Subjects by age, sex, and dementia status. Because of
the recruiting requirements of studies included in the database from
which the data set was culled, a disproportionately large number
of younger subjects were available. Nonetheless, at least 15 subjects
are included in each decade, allowing for continuous sampling of
the adult life span.

eTIV and nWBV can be found in the study of Fotenos

et al. (2005) and Buckner et al. (2004). In the present
sample, eTIV showed a trend for a small age effect [R2 =
0.0022, F(1,158) = 0.35, p = .06 in men, and R2 = 0.12,
F(1,254) = 2.98, p = .09 in women] and a strong sex effect
in each of the young, middle-aged, old without dementia,
Figure 4. Plots of automated anatomical measures by age. Each
and old with dementia groups [t(215) = 10.81, t(95) =
point represents a unique subject from a single scanning session.
6.42, t(97) = 5.41, respectively; all p < .001]. Consistent (Top) eTIV: The best-fit line is drawn for individuals without
with our prior work (Buckner et al., 2005), direct eTIV dementia only. (Bottom) nWBV: The best-fit polynomial regression
comparison of age-matched groups with and without de- is drawn for individuals without dementia only.

Table 3. Sample Characteristics of Older Subjects

CDR = 0 CDR = 0.5 CDR = 1 CDR = 2

Number 98 70 28 2
Sex (female/male) 72/26 39/31 19/9 1/1
Age (years) 75.9 ± 9.0 (60–94) 76.4 ± 7.0 (63–92) 77.2 ± 7.5 (62–96) 82 ± 5.7 (78–86)
Education (years) 14.5 ± 2.9 (8–23) 13.8 ± 3.2 (6–20) 12.9 ± 3.2 (7–20) 11 ± 4.2 (8–14)
MMSE score 29.0 ± 1.2 (25–30) 25.6 ± 3.5 (14–30) 21.7 ± 3.8 (15–29) 15.0 ± 0.0 (15)
Prescriptions (n) 2.8 ± 2.1 (0–8) 2.5 ± 2.0 (0–8) 2.2 ± 2.0 (0–6) 1.5 ± 0.7 (1–2)
Systolic BP (mmHg) 134.2 ± 19.2 (100–192) 142.6 ± 20.2 (104–188) 146.0 ± 25.6 (90–192) 148.0 ± 11.3 (140–156)
Diastolic BP (mmHg) 71.9 ± 10.3 (40–96) 72.7 ± 9.4 (59–90) 74.5 ± 10.6 (60–96) 88.0 ± 19.8 (74–102)
Reported HBP (%) 42.7 43.2 46.4 50.0
Diabetes (%) 9.1 11.6 17.2 0.0

The sample consisted of 198 individuals (98 without dementia and 100 with AD). Values given are mean ± SD. Values in parentheses represent the
range. Full clinical data were not available for 23 subjects. Compared with the adults without dementia, the older adults with dementia had lower
scores on the MMSE [t(194) = 10.66, p < .001] and slightly fewer years of education [t(194) = 2.60, p < .05].
MMSE scores range from 30 (best) to 0 (worst).
HBP = high blood pressure.

Marcus et al. 1503

mentia failed to reach significance for either the men
[t(76) = 0.08, p = .94] or women [t(153) = 1.18, p = .24].
Consistent with the literature, nWBV showed a signif-
icant decrease in individuals without dementia across
the adult life span (18–95 years) [R2 = 0.75, F(2,313) =
485.7, p < .001]. Adding a quadratic term improved the
fit of the model [R2 = 0.78, F(2,313) = 557.1, p < .001).
The decrease in nWBV with age was significant even in
the 18- to 30-year-old group [R2 = 0.048, F(2,138) = 3.9,

Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021

p < .05]. Taken together, these results show brain vol-
ume loss begins early in life and accelerates in advanced
aging. Individuals with AD exhibited marked nWBV
reduction disproportionate to age. Analysis of variance
with age, sex, and dementia status as covariates was sig-
nificant [R2 = 0.44, F(194,3) = 50.4, p < .001], with main
effects for age and dementia status ( p < .001) and a
trend for an effect of sex ( p = .06).

Reliability Scans
To provide a benchmark for determining the reliability
of analytic procedures, 20 subjects were scanned on two
separate occasions with a mean delay of 20.55 days
(range = 1–89 days). These subjects did not have de-
mentia and ranged in age from 19 to 29 years. Because
of the short latency between scan sessions, differences
in the images and measures derived from them are un-
likely to be attributable to anatomical change and in-
stead can be attributed to sources of noise (e.g., head
positioning, day-to-day scanner variability) and instability
in analytic procedures. Figure 5 shows eTIV (top) and
nWBV (bottom) from the first session versus the second
session for each subject in the reliability data set. The
mean absolute percent difference was 0.54% for eTIV
and 0.44% for nWBV. The coefficients of variation (CVs,
the standard deviation of the difference between test and
retest volumes divided by the overall mean, expressed
in percent) were 0.47% for eTIV and 0.44% for nWBV.

Obtaining and Using the Data Figure 5. Plots of automated anatomical measures for reliability
scans. Each point represents the values of the anatomical measure
OASIS data can be obtained at www.oasis-brains.org. Re- calculated from images obtained from the same subject without
quests for DVD distributions of the data can be sub- dementia on two separate scan sessions separated by a short
interval (<90 days). The reliability of eTIV (top, r = 0.99) and
mitted at the Web site. For convenience, we also provide nWBV (bottom, r = 0.96) are high.
the data using the open-source Extensible Neuroimaging
Archive Toolkit (Marcus, Olsen, Ramaratnam, & Buckner,
2007). It provides tools to search, visualize, and down- sion, the following image files are included in the
load the data. Before downloading and requesting distribution: three to four individual scan images; an
data, users are asked to abide by the OASIS Data Usage image in which the individual scans have been aligned
Agreement. and coregistered; an image that has been gain-field-
OASIS data are distributed in GNU zip archive files, corrected and registered to the atlas of Talairach and
which can be uncompressed using freely available soft- Tournoux (1988) (henceforth T88); a masked T88 image
ware. All images are distributed in Analyze 7.5 format, in which the intensity of all nonbrain voxels has been set
which can be visualized and processed using a number to zero; and a segmented T88 image in which each voxel
of commercial and open source applications, including has been labeled as gray matter, white matter, or
Neurolens, ImageJ, and MRIcro. For each imaging ses- cerebral spinal fluid. Demographic, clinical, and derived

1504 Journal of Cognitive Neuroscience Volume 19, Number 9

Table 4. Measures Included in the Data Set to reliably prevent the recognition of particular individ-
Age Age at time of image uals (Bischoff-Grethe et al., in press). Here, we have ad-
acquisition (years) ditionally shown that these methods do not significantly
impact typical calculations of anatomical measures such
Sex Sex (male or female) as intracranial volume and whole-brain volume.
Education Years of education The specific anatomical measures included here, eTIV
Socioeconomic status Assessed by the Hollingshead Index
and nWBV, are illustrative of approaches commonly
of Social Position and classified used to analyze anatomical characteristics of the brain
into categories from 1 (highest in MRI images, particularly in relation to aging. Unsur-

Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021

status) to 5 (lowest status) prisingly, our findings are in agreement with those de-
(Hollingshead, 1957) scribed in previous studies using portions of these data
before defacing (Fotenos et al., 2005; Buckner et al.,
MMSE score Ranges from 0 (worst) to 30 (best)
(Folstein, Folstein, & McHugh, 2004). In particular, eTIV was shown to differ little with
1975) age and dementia status. Because the atlas-based meth-
ods used to generate eTIV are entirely automated, eTIV
CDR scale 0 = no dementia, 0.5 = very mild is an attractive alternative to laborious manual TIV mea-
AD, 1 = mild AD, 2 = moderate
sures for correcting for head size in studies of aging and
AD (Morris, 1993)
dementia. nWBV, on the other hand, was shown to de-
Atlas scaling factor Computed scaling factor (unitless) cline across the adult life span with acceleration in ad-
that transforms native-space brain vanced aging, consistent with findings from a broad
and skull to the atlas target (i.e., range of studies (see DeCarli et al., 2005; Fotenos et al.,
the determinant of the transform
2005; Sowell, Thompson, & Toga, 2004; Durston, 2003;
matrix) (Buckner et al., 2004)
Blatter et al., 1995, for reviews). Marked volume loss was
eTIV Estimated total intracranial volume observed in AD.
(cm3) (Buckner et al., 2004) This distribution also includes data from 20 individ-
nWBV Expressed as the percent of all uals who were scanned on two occasions separated by a
voxels in the atlas-masked image short interval. Because little real change in anatomical
that are labeled as gray or white structure is likely to have occurred between the two
matter by the automated tissue acquisitions, these images are useful for estimating the
segmentation process (Fotenos reliability of analytic methods. The measures of eTIV
et al., 2005) and nWBV described here provide a reference point for
the amount of error seen with established approaches.
However, because these scans were all obtained from
young healthy individuals, reliability may be overesti-
imaging measures (Table 4) are available in XML and mated relative to a sample of older subjects with de-
spreadsheet formats. Additional details of the directory mentia who are more prone to movement and other
structure, file naming scheme, and image characteristics artifacts.
can be found at www.oasis-brains.org/pdf/oasis_cross- Although other publicly available neuroimaging data
sectional_facts.pdf. sets are available, most notably those provided by the
Alzheimer’s Disease Neuroimaging Initiative (ADNI), the
present data set contributes complimentary features.
First, ADNI targets mild cognitive impairment and AD
DISCUSSION
and, by design, includes a limited number of older adults
The present data set includes T1-weighted MRI data without dementia and no individuals before middle age.
from 416 individuals aged 18 to 96 years, including 100 The present data set targets the full adult life span and,
individuals clinically diagnosed with AD. Multiple acqui- therefore, allows exploration of aging in addition to
sitions are included for each subject, allowing extremely effects of AD. Second, ADNI is a multicenter study that
high contrast properties after image averaging. These will be extremely valuable for developing methods that
data are amenable to a range of analysis procedures. can pool across sites. There will not, however, be a large
The data have been de-identified, carefully screened for quantity of matched data sets from the same scanner/
image quality, and postprocessed to generate common sequence type. The present data includes more than
anatomical measures. These data are available under a 400 subjects with data collected on the identical plat-
liberal usage policy that allows free access to all inter- form with identical sequences. Finally, in the present
ested parties. data set, multiple acquisitions were acquired for each
To accommodate public release of the data, the orig- subject within session to allow substantial signal aver-
inal images were processed to remove facial features. aging. This is not typically done, and this choice was
The methods used to accomplish this have been shown made to optimize our data for exploration of automated

Marcus et al. 1505

computational analyses, including measures of cortical old, and demented adults using automated atlas-based
thickness and fine demarcation of subcortical regions. head size normalization: Reliability and validation against
manual measurement of total intracranial volume.
The result is 416 scanning sessions, each with high- Neuroimage, 23, 724–738.
contrast imaging data that are amenable to a wide range Buckner, R. L., Quinn, B. T., Salat, D. H., Head, D., Snyder,
of computational analyses. A. Z., Busa, E., et al. (in press). Evidence for distinct effects
A number of additional data sets are planned as part of aging and Alzheimer’s disease on the hippocampus.
of the OASIS project. These will include longitudinal Journal of Neuroscience.
Buckner, R. L., Snyder, A. Z., Shannon, B. J., LaRossa, G.,
acquisitions of individuals with and without dementia, Sachs, R., Fotenos, A. F., et al. (2005). Molecular, structural,
acquisitions from subjects obtained on multiple different and functional characterization of Alzheimer’s disease:

Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021

scanners, functional MRI of cognition, and resting-state Evidence for a relationship between default activity,
functional MRI for noise analysis and analysis of sponta- amyloid, and memory. Journal of Neuroscience, 25,
neous low-frequency signals. Following the principles 7709–7717.
Burns, J. M., Church, J. A., Johnson, D. K., Xiong, C.,
established with the initial release, these sets will be Marcus, D., Fotenos, A. F., et al. (2005). White matter
carefully prepared, vetted, and described. lesions are prevalent but differentially related with
cognition in aging and early Alzheimer disease. Archives
of Neurology, 62, 1870–1876.
Acknowledgments Coffey, C. E., Lucke, J. F., Saxton, J. A., Ratcliff, G., Unitas,
L. J., Billig, B., et al. (1998). Sex differences in brain aging:
We thank the Washington University ADRC and the Conte Cen- A quantitative magnetic resonance imaging study. Archives
ter for clinical assistance and participant recruitment; Elizabeth of Neurology, 55, 169–179.
Grant for assistance with data preparation; Susan Larson, Amy Coffey, C. E., Saxton, J. A., Ratcliff, G., Bryan, R. N., & Lucke,
Sanders, Laura Williams, and Glenn Foster for assistance with J. F. (1999). Relation of education to brain size in normal
MRI data collection; Avi Snyder for development of analytic aging: Implications for the reserve hypothesis. Neurology,
techniques; Anthony Fotenos for statistical contributions; and 53, 189–196.
Tim Olsen and Mohana Ramaratnam for development of data- Csernansky, J. G., Wang, L., Joshi, S. C., Ratnanather, J. T.,
base and web tools. Anders Dale assisted with the original & Miller, M. I. (2004). Computational anatomy and
selection of imaging parameters. The acquisition of these data neuropsychiatric disease: Probabilistic assessment of
and support for data analysis and management were provided variation and statistical inference of group difference,
by NIH grants P50 AG05681, P01 AG03991, P20 MH071616, hemispheric asymmetry, and time-dependent change.
RR14075, RR 16594, BIRN002, the Alzheimer’s Association, the Neuroimage, 23 (Suppl. 1), S56–S68.
James S. McDonnell Foundation, the Mental Illness and Neuro- DeCarli, C., Massaro, J., Harvey, D., Hald, J., Tullberg, M.,
science Discovery Institute, and the Howard Hughes Medical Au, R., et al. (2005). Measures of brain morphology and
Institute. infarction in the Framingham Heart Study: Establishing
Reprint requests should be sent to Daniel S. Marcus, Depart- what is normal. Neurobiology of Aging, 26, 491–510.
ment of Radiology, Washington University School of Medicine, Desikan, R. S., Segonne, F., Fischl, B., Quinn, B. T.,
Campus Box 8225, 4525 Scott Avenue, St. Louis, MO 63110, or Dickerson, B. C., Blacker, D., et al. (2006). An automated
via e-mail: dmarcus@wustl.edu. labeling system for subdividing the human cerebral
cortex on MRI scans into gyral based regions of interest.
Neuroimage, 31, 968–980.
Durston, S. (2003). A review of the biological bases of
REFERENCES ADHD: What have we learned from imaging studies?
Mental Retardation and Developmental Disabilities
Berg, L., Hughes, C. P., Coben, L. A., Danziger, W. L., Martin, Research Reviews, 9, 184–195.
R. L., & Knesevich, J. (1982). Mild senile dementia of Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M.,
Alzheimer type: Research diagnostic criteria, recruitment, Haselgrove, C., et al. (2002). Whole brain segmentation:
and description of a study population. Journal of Automated labeling of neuroanatomical structures in the
Neurology, Neurosurgery and Psychiatry, 45, 962–968. human brain. Neuron, 33, 341–355.
Berg, L., McKeel, D. W., Jr., Miller, J. P., Storandt, M., Rubin, Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975).
E. H., Morris, J. C., et al. (1998). Clinicopathologic studies ‘‘Mini-Mental State.’’ A practical method for grading the
in cognitively healthy aging and Alzheimer’s disease: cognitive state of patients for the clinician. Journal of
Relation of histologic markers to dementia severity, age, Psychiatric Research, 12, 189–198.
sex, and apolipoprotein E genotype. Archives of Neurology, Fotenos, A. F., Mintun, M. A., Snyder, A. Z., Morris, J. C.,
55, 326–335. & Buckner, R. L. (in press). Brain volume decline in aging:
Bischoff-Grethe, A., Ozyurt, I. B., Busa, E., Quinn, B. T., Evidence for a relationship between socioeconomic status,
Fennema-Notestine, C., Clark, C. P., et al. (in press). preclinical AD, and reserve. Journal of Neuroscience.
A technique for the deidentification of structural brain Fotenos, A. F., Snyder, A. Z., Girton, L. E., Morris, J. C., &
MR images. Human Brain Mapping. Buckner, R. L. (2005). Normative estimates of cross-sectional
Blatter, D. D., Bigler, E. D., Gale, S. D., Johnson, S. C., and longitudinal brain volume decline in aging and AD.
Anderson, C. V., Burnett, B. M., et al. (1995). Quantitative Neurology, 64, 1032–1039.
volumetric analysis of brain MR: Normative database Head, D., Snyder, A. Z., Girton, L. E., Morris, J. C., &
spanning 5 decades of life. American Journal of Buckner, R. L. (2005). Frontal–hippocampal double
Neuroradiology, 16, 241–251. dissociation between normal aging and Alzheimer’s
Buckner, R. L., Head, D., Parker, J., Fotenos, A. F., disease. Cerebral Cortex, 15, 732–739.
Marcus, D., Morris, J. C., et al. (2004). A unified approach Hollingshead, A. (1957). Two factor index of social position.
for morphometric and functional data analysis in young, New Haven, CT: Yale University Press.

1506 Journal of Cognitive Neuroscience Volume 19, Number 9

Jack, C. R., Jr., Petersen, R. C., O’Brien, P. C., & Tangalos, and executive functions. Behavioral Neuroscience,
E. G. (1992). MR-based hippocampal volumetry in the 117, 1169–1180.
diagnosis of Alzheimer’s disease. Neurology, 42, 183–188. Raz, N., Williamson, A., Gunning-Dixon, F., Head, D., &
Jack, C. R., Jr., Petersen, R. C., Xu, Y. C., Waring, S. C., Acker, J. D. (2000). Neuroanatomical and cognitive
O’Brien, P. C., Tangalos, E. G., et al. (1997). Medial correlates of adult age differences in acquisition of a
temporal atrophy on MRI in normal aging and very perceptual–motor skill. Microscopy Research and
mild Alzheimer’s disease. Neurology, 49, 786–794. Techniques, 51, 85–93.
Killiany, R. J., Moss, M. B., Albert, M. S., Sandor, T., Tieman, J., Salat, D. H., Buckner, R. L., Snyder, A. Z., Greve, D. N.,
& Jolesz, F. (1993). Temporal lobe regions on magnetic Desikan, R. S., Busa, E., et al. (2004). Thinning of the
resonance imaging identify patients with early Alzheimer’s cerebral cortex in aging. Cerebral Cortex, 14, 721–730.
disease. Archives of Neurology, 50, 949–954. Scheltens, P., Leys, D., Barkhof, F., Huglo, D., Weinstein,

Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/19/9/1498/1756878/jocn.2007.19.9.1498.pdf by guest on 18 May 2021

Koga, H., Yuzuriha, T., Yao, H., Endo, K., Hiejima, S., H. C., Vermersch, P., et al. (1992). Atrophy of medial
Takashima, Y., et al. (2002). Quantitative MRI findings temporal lobes on MRI in ‘‘probable’’ Alzheimer’s
and cognitive impairment among community dwelling disease and normal ageing: Diagnostic value and
elderly subjects. Journal of Neurology, Neurosurgery neuropsychological correlates. Journal of Neurology,
and Psychiatry, 72, 737–741. Neurosurgery and Psychiatry, 55, 967–972.
Marcus, D. S., Olsen, T., Ramaratnam, M., & Buckner, R. L. Snyder, A. Z. (1996). Difference image vs ratio image error
(2007). The extensible neuroimaging archive toolkit function forms in PET–PET realignment. In D. Bailey &
(XNAT): An informatics platform for managing, exploring, T. Jones (Eds.), Quantification of brain function using
and sharing neuroimaging data. Neuroinformatics, 5, 11–34. PET (pp. 131–137). San Diego, CA: Academic Press.
McKhann, G., Drachman, D., Folstein, M., Katzman, R., Sowell, E. R., Thompson, P. M., Leonard, C. M., Welcome,
Price, D., & Stadlan, E. M. (1984). Clinical diagnosis of S. E., Kan, E., & Toga, A. W. (2004). Longitudinal mapping
Alzheimer’s disease: Report of the NINCDS-ADRDA Work of cortical thickness and brain growth in normal children.
Group under the auspices of Department of Health and Journal of Neuroscience, 24, 8223–8231.
Human Services Task Force on Alzheimer’s Disease. Sowell, E. R., Thompson, P. M., & Toga, A. W. (2004).
Neurology, 34, 939–944. Mapping changes in the human cortex throughout the
Morris, J. C. (1993). The Clinical Dementia Rating (CDR): span of life. Neuroscientist, 10, 372–392.
Current version and scoring rules. Neurology, 43, Storandt, M., Grant, E. A., Miller, J. P., & Morris, J. C. (2006).
2412–2414. Longitudinal course and neuropathologic outcomes in
Morris, J. C., Storandt, M., Miller, J. P., McKeel, D. W., original vs revised MCI and in pre-MCI. Neurology, 67,
Price, J. L., Rubin, E. H., et al. (2001). Mild cognitive 467–473.
impairment represents early-stage Alzheimer disease. Styner, M., Brechbuhler, C., Szekely, G., & Gerig, G.
Archives of Neurology, 58, 397–405. (2000). Parametric estimate of intensity inhomogeneities
Mugler, J. P., III, & Brookeman, J. R. (1990). Three-dimensional applied to MRI. IEEE Transactions on Medical Imaging,
magnetization-prepared rapid gradient-echo imaging (3D 19, 153–165.
MP RAGE). Magnetic Resonance in Medicine, 15, 152–157. Talairach, J., & Tournoux, P. (1988). Co-planar stereotaxic
Raz, N., Gunning, F. M., Head, D., Dupuis, J. H., McQuain, J., atlas of the human brain: An approach to medical
Briggs, S. D., et al. (1997). Selective aging of the human cerebral imaging. New York: Thieme.
cerebral cortex observed in vivo: Differential vulnerability Zhang, Y., Brady, M., & Smith, S. (2001). Segmentation of
of the prefrontal gray matter. Cerebral Cortex, 7, 268–282. brain MR images through a hidden Markov random field
Raz, N., Rodrigue, K. M., & Acker, J. D. (2003). Hypertension model and the expectation–maximization algorithm.
and the brain: Vulnerability of the prefrontal regions IEEE Transactions on Medical Imaging, 20, 45–57.

Marcus et al. 1507