UC San Diego
UC San Diego Previously Published Works
Title
Reduced cerebral grey matter observed in alcoholics using magnetic resonance imaging.
Permalink
https://escholarship.org/uc/item/6gv300dp
Journal
Alcoholism, clinical and experimental research, 15(3)
ISSN
0145-6008
Authors
Jernigan, TL
Butters, N
DiTraglia, G
et al.
Publication Date
1991-06-01
DOI
10.1111/j.1530-0277.1991.tb00540.x
Peer reviewed
eScholarship.org Powered by the California Digital Library
University of California
0145-6008/9 1/1503-04 I8$3.00/0
ALSOHOLISM:
Cl.INI(AI
AN11
EXPCRIMENTAL
RI.St.AKC'I1
Vol.
15,
No.
3
May/June
1991
Reduced Cerebral Grey Matter Observed
in
Alcoholics
Using Magnetic Resonance Imaging
Terry
L.
Jernigan, Nelson Butters, Gina DiTraglia, Kimberly Schafer, Tom Smith, Michael Irwin, lgor Grant,
Marc Schuckit, and Laird
S.
Cermak
Twenty-eight chronic alcoholics and
36
age- and sex-matched non-
alcoholic controls were examined with magnetic resonance imaging
and brain morphometric analyses. Results confirmed large increases
in subarachnoid cerebrospinal fluid (CSF) volume and mild ventric-
ular enlargement in the alcoholics and revealed associated volume
reductions of localized cortical and subcortical cerebral structures.
Volume losses in the diencephalon, the caudate nucleus, dorsolat-
era1 frontal and parietal cortex, and mesial temporal lobe structures
were the most prominent. Significant correlations between incre-
ments in cortical and ventricular CSF and decrements in the volume
of cortical and subcortical grey matter were noted. Although there
was little evidence for relationships between performance on neu-
ropsychological tests and volume of grey matter structures, signifi-
cant correlations between some cognitive measures and subcortical
and cortical fluid volumes were found. The parallels between this
pattern
of
affected structures and recent neuropathological findings
are discussed.
Key Words:
MRI,
Alcoholism, Cerebral Atrophy.
REVIOUS NEURORADIOLOGICAL comparisons
P
of
chronic alcoholics and nonalcoholic controls have
consistently demonstrated cerebrospinal (CSF) increases
in the brains of the alcoholics, especially in the subarach-
noid spaces.'-'" These changes would seem to support
neuropathological reports
of
reduced brain weight and
brain volume in alcoholics"-'3 and perhaps even the gen-
eral notion that ethanol is a neurotoxic agent affecting a
number of brain regions important for higher cognitive
functions.'4-" However, this interpretation of these neu-
roradiological findings has been challenged by reports that
these increases in CSF are at least partially reversible with
prolonged and may actually reflect a tran-
sitory brain overhydration during detoxification.2"26 In
From the Psychology Service, Department of Veterans Affairs Medical
Center, Sun Diego, California (T.L.J., N.B., G.D.T.,
K.S.);
Psychiatry
Service, Department
of
Veterans Affairs Medical Center, San Diego,
California (T.S.,
M.I.,
I.G.,
M.S.); Psychiatry Departrnerlt, University of
California
San
Diego, School
of
Medicine, San Diego, California (T.L.J.,
N.B.,
M.I.,
I.G.,
M.S.);
Psychology Service, Department of Veterans
Affairs Medical Center, Boston, Massachusetts
(L.S.
C.);
Neurology De-
partment, Boston University School
of
Medicine, Boston, Massachusep
(L.S.C.).
Received for publication July 17, 1990; accepted November
14,
1990.
This research wa.c
supported
by
funds
fvom the Department of Veterans
Affairs Medical Research Service and National Institute
on
Alcohol Abuse
and Alcoholism grant AA-00187 to Boston University.
Reprint requests: Dr. Terry L. Jernigan, University of Cali$ornia-San
Diego, Department of Psychiatry, 0631-P, 9500 Gilman Drive, La Jolla,
CA 92093-0631.
Copyright
0
1991 by The Research Society
on
Alcoholism.
41
a
addition, the failure, in some instances, of these increases
in CSF to be consistently and highly correlated with
neuropsychological deficits after the effect of age has been
removed has raised further doubts concerning the per-
manent pathological implications of the observed CSF
increments.
1,2.5
With the development of quantitative (morphometric)
techniques for evaluating the volume of grey matter struc-
tures seen on magnetic resonance (MR) images, it is now
possible to ascertain more precisely the implications
of
the CSF increases in alcoholics' CT scans and pneumoen-
cephalograms. In an initial study, Jernigan et al.27 com-
pared CSF increases and grey matter reductions in eight
alcoholics with Korsakoffs Syndrome,
12
age-matched
nonamnesic alcoholics, and
13
age-matched nonalcoholic
controls. Both alcoholic groups had significant cortical
and subcortical grey matter losses in association with CSF
increases, but these changes were more extensive in the
amnesic alcoholics. While grey matter reductions in the
diencephalon and superior fronto-parietal cortices char-
acterized alcoholism per se, the Korsakoff patients were
distinguished by unusual volume losses in the anterior
diencephalon, mesial temporal, and orbitofrontal regions.
Since recent neuropathological
report^'^-'^^^^-^^
have noted
abnormalities in cortical, diencephalic, basal forebrain,
and hippocampal structures of alcoholics, these volume
decrements on MR may actually be in vivo indices of
significant cellular changes in grey matter.
In the present study, the findings
of
Jernigan et al.27 are
extended to a larger and younger population of nonam-
nesic alcoholics who have been administered an extensive
neuropsychological examination. Besides a detailed eval-
uation of the relationship between CSF increases and
specific grey matter losses and white matter signal hyper-
intensities, correlations between cognitive losses and var-
ious MR indices are also assessed. It was anticipated that
the greater resolution of MR combined with morphomet-
ric techniques would result in higher brain-behavior cor-
relations than usually reported in previous CT scan stud-
ies. Furthermore, increments in sulcal and ventricular CSF
were expected to be correlated with volume reductions in
association cortices and subcortical nuclei, respectively.
If
both
of
these predictions were realized, they would
strengthen the notion that the CSF increments noted in
neuroradiological reports of alcoholics do indeed reflect
significant changes in various grey matter structures.
AlcoholClin
ExpRes,
Vol
15,
No
3,
1991: pp418-427
REDUCED
CEREBRAL
GREY
MATTER
419
MATERIALS AND METHODS
Subjects
Twenty-eight alcoholic males (mean age
=
49.5 years, SD
=
9.9) were
subjects in this study. These patients had undergone detoxification prior
to admission to the San Diego Department of Veterans Affairs Medical
Center’s Alcohol Treatment Program (ATP), a 28-day program for
alcoholism counseling and treatment. Using the Alcohol Research Center
Intake Interview,” data on drinking and medical histories were obtained
from each patient and at least one resource person, such as a close friend
or
family member, and the diagnosis of alcohol abuse
or
dependence
was documented using DSM-II13’ criteria. Individuals were excluded
from this study
if
they had a history of overt liver (e.g., cirrhosis,
jaundice), metabolic (eg, diabetes), vascular (e.g., coronary artery dis-
ease),
or
neurologic (e.g., head injury, encephalitis, epilepsy) disorders.
Patients with a history of drug abuse
or
of major psychiatric illness (eg,
schizophrenia, PTSD, bipolar affective disorder) predating the onset of
alcoholism were also screened from the study. The alcoholics’ years of
alcohol abuse and their mean daily consumption of ethanol for the 3-
month period prior to admission to the ATP are shown in Table
1.
Ten
of the older alcoholic patients in the present study were also subjects in
the previous report comparing MRI indices of alcoholics with and
without Korsakoffs Syndrome.”
All
28
alcoholics were administered a neuropsychological examination
comprised primarily of tests known
to
be affected in recently detoxified
patients.34 The Vocabulary test (WAIS-R), Trails A and
B,
Digit Symbol
(WAIS),
and Visual Search Test3’ were administered within
48
hr of
Table
1.
Mean Age, Education, Drinking Variables, and Neuropsychological Test
Scores for Alcoholic
(N
=
28)
and Nonalcoholic Control
(N
=
15)
Subjects
Alcoholics Controls
P‘
Age (years)
Education (years)
Years of alcoholism
Daily ethanol consump-
tion
in
3-months prior
to
admission (mean
number
of
drinks
per
day)
Vocabulary (WAIS-R
scaled score)
Trails
A
(seconds)
Trails
B
(seconds)
Digit
symbol (WAIS
scaled score)
Visual search (seconds)
Category Test (total er-
rors)
Stroop Test (Card
3:
number correct)
Story Recall-immediate
(number correct: trial
1)
Story Recall-trials-to-
criterion
Story Recall-delayed
(number correct)
RAVLT-immediate re-
call (number correct:
trial
1)
RAVLT-immediate re-
call (total correct: trials
1-5)
RAVLT-delayed recall
(number correct)
49
5
(9
9)
13
9
(1
9)
12
0
(8
8)
14
9
(10
0)
9
8
(2
0)
31
8
(10
3)
107 6
(48
1)
9 0
(2
3)
164
3
(95
1)
64
0
(27
3)$
33 3
(8
I)$
12
4
(4
2)
2 1 (1 1)
15 5
(2
8)
5
6
(1
5)
43
0 (8
6)
9 0
(2
6)
50.4
(10.0)
15.1 (1.6)
0.4
(0.9)t
12.1 (1.9)
25.0
(6.4)
66.9 (22.6)
11.6 (3.0)
126.2
(48.9)
35.6 (17.6)
40.9
(7.7)
15.7 (2.8)
1.4 (0.5)
16.5
(3.1)
7.7 (2.2)
49.7
(8.8)
9.8
(3.2)
NS
<0.05
<0.01
co.01
<0.06
co.01
<0.01
NS
<0.01
<0.02
<0.02
NS
NS
<0.01
<0.06
NS
’
Age, education and daily ethanol consumption were assessed
with
two-tail
t-
tests, neuropsychological test scores
with
analysis
of
covariance controlling for
education differences
7
Daily ethanol consumption for the
3
months prior to
the
interview
was available
for only
13
of
the
15
nonalcoholic controls
$
The Category Test and Stroop
Test
were administered
to
only
27
alcoholics
because one subject was colorblind
admission. The booklet form of the Category Test,36 the Stroop
Color
&
Word Test,” the Rey Auditory Verbal Learning Test (RAVLT),38 and
Story Recall (Story
1
of the Logical Memory test from the Wechsler
Memory Scale-Revised3’) were administered 3 to 4 weeks following
admission.
For
Story Recall, the number of correctly recalled ideas on
the first test trial, trials to reach a predetermined learning criterion, and
the number of correctly recalled ideas after a 30-min delay were recorded.
The 36 nonalcoholic male controls (mean age
=
48.0 years, SD
=
10.1) were recruited from the community by newspaper advertisements.
All
of these control subjects were screened
for
a history of alcohol abuse,
alcoholism, drug abuse, and the same medical and psychiatric disorders
described for the alcoholic subjects. Thirteen of the older nonalcoholic
controls also participated in Jernigan et al.’s3’ previous MRI study of
alcoholic Korsakoff
s
Syndrome. Of the 36 nonalcoholic controls, 15
were administered the same neuropsychological examination as the
alcoholics.
Table
1
shows the mean age, education, drinking histories (i.e., years
of alcoholism, mean daily consumption for the 3-month period prior to
admission), and neuropsychological test scores for the 28 alcoholics and
15 controls who were administered psychological tests. Since the two
groups’ difference in education was significant
(p
<
0.05),
analyses of
covariance controlling for education were used to compare their perform-
ances on the neuropsychological tests.
As
expected, the alcoholics were
impaired on most of the test scores, and thus, appeared similar on the
basis of their cognitive deficits, as well as their drinking history, to
patients reported in other neuropsychological and neuroradiological stud-
ies,
1.34.40-43
Imuging
Protocol
MR was performed with a 1.5-T super-conducting magnet (Signa;
General Electric, Milwaukee), at the UCSD/AMI Magnetic Resonance
Institute. A standard protocol was used for the acquisition of MR brain
images, and the images were analyzed in the Brain Image Analysis
Laboratory of the Department of Psychiatry, UCSD. Proton-density
weighted (PDW) and Tz-weighted (TZW) images (Fig.
I)
were obtained
simultaneously
for
each section, using an asymmetrical, multiple-echo
sequence (TR
=
2000
msec, TE
=
25,
70
msec) to obtain images of the
entire brain in the axial plane. Section thickness was
5
mm with a 2.5
mm gap between successive sections in all instances. A 256
X
256 matrix
and 24 cm field of view were used.
No
sedation was administered for the
examinations.
For
the following discussion of image analysis, the term
pixel will be used to refer to a single picture element
(or
signal value)
from the image matrix. The term voxel will be used to refer to the
corresponding 3-dimensional volume from which the signal value for a
pixel is taken.
Image
Analysis
The visual identification of specific structures in MR images is possible
because of the tissue contrast between the grey matter structures and the
surrounding white matter
or
CSF. However, measurements of volumes
of cerebral structures must overcome several problems. First, because of
partial-voluming
of
grey matter with white matter,
or
CSF
(or
all three)
at the edges of structures, sharply defined edges are not always present.
This allows considerable scope for variability in subjective determinations
of such boundaries when, for example, tracing methods are used, leading
to measurement unreliability in the computed volumes.
Visual determination of specific cortical structures on MRI presents
additional challenges and depends upon the presence of visible
gross
morphologic features relative to which the boundaries of the cortical
regions can be defined. Standard regional divisions for the cortex are
based to a large extent on cortical gyral patterns, but the accurate
localization of particular gyri
or
sulci, throughout a series
of
images, is
often impossible. Furthermore, some boundaries, such as that between
posterior temporal and inferior parietal cortex, are not clearly defined in
gross
morphological terms. Also, even when attempts are made to
420
JERNIGAN
ET
AL.
standardize head positioning, rotation of the head (relative to the imaging
plane) occurs in all three planes. This is especially true with unsedated
subjects who must be sufficiently comfortable in position to avoid
movement during the imaging session. Careful inspection indicates that
relatively small rotations substantially change the appearance
of
brain
structures in the image plane, further complicating their visual identifi-
cation. Thus, manually tracing the structures in the sections where they
are best visualized often leads to inaccurate volume and asymmetry
assessments. The techniques described below are designed to address
these difficulties.
To facilitate and standardize the determination of structural edges,
our
method involves a semi-automated classification of all pixels in the
images
on
the basis of their signal characteristics
on
the two original
image matrices for each section.
A
detailed description of the basic image
analysis method has been rep0rted.4~ Only
a
brief summary is provided
here: Each axial image is first digitally filtered to reduce the signal drift
across the image due to magnetic field and gradient inhomogeneities.
Information in the two images for each axial section is then combined
to
best
distinguish the different tissues in the image. For each section
imaged,
a
computed matrix is produced.
In
this matrix, voxels are
classified as most resembling (in signal strength) grey matter, white
matter,
CSF,
or
signal hyperintensities (tissue abnormalities). The full
series of axial images is analyzed, beginning at the bottom ofthe cerebellar
hemispheres and extending through the vertex.
Further manipulations to derive the specific structural measures
for
the present study were then made using these “pixel-classified’’ images.
Trained operators, blind to any subject characteristics, used a stylus-
controlled cursor
on
the displayed images to manually separate infraten-
tonal from supratentorial areas, left from right hemispheres, and the
cortical from subcortical regions of the supratentorial cranium. Thus,
separate estimates of the four classes of pixels were made for these areas.
Dejnition
of
Subcortical
Structures
To delineate subcortical structures, the operators circumscribed pixels
classified as subcortical grey matter that were visually determined
to
be
in caudate nuclei, lenticular nuclei, and diencephalic grey matter struc-
tures (including mammillary bodies, other hypothalamic grey, septa1
nuclei, and thalamus). They did not trace the edges of the structures
(since the pixel classification defined the transition from grey
to
surround-
Fig.
1.
Representative images from the stand-
ard protocol. A, Axial section,
SE
2000/25
(PDW
in text).
B,
Axial section, SE 2000/70
(T2W
in text). Sections are 5 mm thick, matrix
256
x
256, with 2.5 mm gaps between images.
A field of view of 24 cm
was
used.
ing white matter), but defined polygons that included all grey matter
pixels within the structures, and excluded those grey matter pixels
associated with other structures.
In
some cases, when the subcortical
nuclei were contiguous with other areas classified as grey but clearly not
in the structures, boundaries were manually constructed using the filmed
images as
a
guide.
Definition
of
Cortical Regions
To
define anatomically consistent cortical regions,
a
method was
adopted for making subdivisions of the supratentorial cranium relative
to
the centromedial structural midline and two consistently identifiable
points: the most anterior midline point in the genu and the most posterior
midline point in the splenium of the corpus callosum. By calculating
rotation angles using these landmarks, it was possible
to
perform
a
3-
dimensional rotation of the images, thus correcting each individual’s
image data for rotation out of the optimal imaging plane. Regions could
then
be
constructed which resulted in highly consistent placement of
regional boundaries relative to gross anatomical landmarks.
The two corpus callosum points were considered to lie in the true
midsagittal plane. The orientation of this plane was then determined by
computing
a
regression line through a series of visually-selected brainstem
midline points
on
different sections. The division of the cerebrum was
based
on
two major planes (see Fig.
2):
an
axial plane,
which is perpen-
dicular in orientation to the midsagittal plane and passes through the
two corpus callosum points, and
a
coronal plane,
which is defined as
perpendicular to the first plane and which passes through the midpoint
between the two corpus callosum points. By computing new coordinates
for each voxel relative to these planes, each is assigned to one of four
zones: one, inferior to the axial plane and anterior to the coronal plane
(IA); a second, inferior to the axial plane and posterior to the coronal
plane (IP); a third, superior to the axial plane and anterior to the coronal
plane (SA); and a fourth, superior to the axial plane and posterior to the
coronal plane
(SP).
Again, these defined planes are independent of the
image plane, as a 3-dimensional rotation is first applied based
on
the
positions of the landmarks described above. Anterior temporal, orbito-
frontal, and some dorsolateral and mesial frontal cortex lie in the inferior
anterior zone. Posterior temporal and inferior occipital cortex fall in the
inferior posterior zone. Most of the remaining parts of the frontal lobe
fall into the superior anterior zone, and the superior posterior zone
REDUCED CEREBRAL
GREY
MATTER
42
1
Fig.
2.
Cerebral regions are defined as follows: Points A and B in the corpus
callosum, shown above, are the most anterior midline point in the genu, and the
most
posterior midline point in the splenium, respectively. An axial plane passing
through these two points is defined,
as
shown, perpendicular to the midsagittal
plane. A coronal plane is defined perpendicular
to
the axial plane and passing
through the midpoint between points
A
and B. Thus four cerebral zones are
defined: inferior anterior, inferior posterior, superior anterior, and superior posterior.
Anterior temporal, orbitofrontal, and some dorsolateral and mesial frontal cortex lie
in the inferior anterior zone. Posterior temporal and inferior occipital cortex fall in
the inferior posterior zone.
Most
of the remaining parts of the frontal lobe fall into
the superior anterior zone, and the superior posterior zone contains primarily
parietal and superior occipital cortex.
contains primarily parietal and
a
small portion of the superior occipital
cortex.
To further separate mesial from peripheral cortical regions, an ellip-
soid volume was defined within the supratentorial cranial vault. This
volume constitutes
30%
of the supratentorial volume and has cardinal
dimensions proportional to those of the supratentorial vault (i.e., the z-
axis extent of the ellipsoid is proportional to the maximum z-axis extent
of the supratentorial cranium, the y-axis extent of the ellipsoid to the
maximum y-axis extent, and the x-axis to the maximum x-axis extent).
The ellipsoid is centered slightly behind, but in the same axial plane as,
the origin of the coordinate system described above, at a point
60%
of
the distance from the genu to the splenium reference point along the line
connecting them. The size and center point of the volume were chosen
empirically
so
as to isolate as well as possible the medial cortical surfaces
of the limbic lobe,“5 while excluding the more lateral neocortical surfaces.
The area designated as mesial with this method is shown in green in Fig.
3.
It consistently includes the most posterior parts of the orbital frontal
lobe, the amygdala, the hippocampus and most of the parahippocampal
gyrus, the insula, and most of the cingulate gyrus. The ellipsoid defines
mesial and peripheral zones within each of the four original cerebral
zones described above. A summary of the cortical structures falling into
each of the resulting eight zones is given in Table
2.
The fully processed images are illustrated in Fig.
3.
The images shown
are the actual color-coded digital images produced for one of the subjects.
The different pixel classes are color coded
as
follows: Diencephalic areas
are purple, caudate nuclei are blue, and lenticular nuclei are magenta.
Within the subcortical white matter there are some voxels with signal
values that fall not within the criterion range for white, but within’ the
range of grey matter values (i.e., they demonstrate lengthened
Tz
relative
to other white matter voxels). These voxels have been coded separately
and are shown in Fig.
3
in yellow. The red line running through each
section indicates the position of the coronal dividing plane. This plane
consistently intersects the amygdala and falls between the column of the
fornix and the mammillothalamic tract (See Fig.
3).
At the level of the
mammillary bodies it consistently divides them from the more anterior
hypothalamic grey areas. Because this plane passes through the dience-
phalic grey matter regions and divides the anterior hypothalamic and
septa1 structures (lying anteriorly) from the bulk of the thalamus (lying
posteriorly), the corresponding anterior and posterior diencephalic areas
were examined separately. It should be noted that areas within the
lenticular nucleus containing significant iron deposits, particularly in
globus pallidus, do not meet the signal criteria for grey matter and are
thus not included in this region. Fluid and white matter are shown in
red and black, respectively; however, subcortical and cortical fluid are
measured separately.
Volume of the supratentorial cranium was estimated by summing
supratentorial voxels (including CSF, hyperintensities, and grey and
white matter) over all sections. The subcortical and cortical CSF voxels
were summed over all sections separately to estimate the ventricular and
cortical sulcal volumes. The grey matter voxels within each
of
the
subcortical structures and the cortical grey matter voxels within each of
the eight cerebral zones were summed separately. Eight regional volumes
were also computed by summing all supratentorial voxels (including
CSF, hyperintensities, and grey and white matter) within each region.
All subcortical measures and both CSF measures were expressed as
proportions of the supratentorial cranial volume, and all cortical grey
matter measures as proportions of their respective regional volumes.
Finally, an index of signal alterations in the white matter
was
constructed
by summing voxels within the subcortical white matter regions having
signal characteristics meeting criteria for “grey matter” or for “signal
hyperintensities”, i.e., they had longer
Tz
values. This measure was also
expressed as
a
proportion of the supratentorial cranial volume. It should
be mentioned that while all subjects show some voxels in the subcortical
white matter with longer
T2,
the number of such voxels has been shown
to increase significantly with age” and the presence of dementia.46
Slut
istical
Analysis
Because there are well-established age changes in most, if not all, of
these measures, it is advisable to remove age effects before estimating the
group effects. In
our
previous articles we, and others, have pointed out
that removing age effects from the full sample (including alcoholics)
tends to remove part of the “alcoholism” effect, because older alcoholics
(with longer drinking histories) usually have more atrophy, which is
incorrectly attributed to “aging.” The approach we have used avoids this
problem, and provides an accurate means of removing variance associ-
ated with normal aging even when examining alcoholic patients in whom
age and years of alcoholism are usually confounded. This method also
facilitates comparison of the extent of damage to different structures by
placing the measures on a standard scale.
These measures were converted to age-corrected z-scores using for-
mulae derived from control data in
55
normal volunteers ranging from
30
to
79
years of age.*’ These values, by definition, have an expected
mean of
0
and a standard deviation of
1
in the controls. The group
means presented here are the averages of these z-scores.
The computation of age-adjusted scores is described in greater detail
in earlier
report^.^.'^,^',^^
These values take advantage of the fact that we
have examined a larger group of controls and estimated the changes with
aging of both mean and standard deviation of each measure. On the
basis of this information we have then expressed each subject’s value in
terms of its deviation from the mean of his normal age-peers in the
standard deviation units observed at his age. The coefficients used to
compute the age-adjusted scores are given in the Appendix.
To
ensure
that the full control sample was not in some way unrepresentative of the
alcoholic sample, we directly compare the values in the alcoholics with
the present sample
of
controls, carefully matched to the alcoholics by
age and gender. Differences between the means of the z-scores for the
alcoholic patients and those of their age- and sex-matched controls were
significance tested with Student’s
t
tests. Degree of correlation between
measures was estimated with the Pearson Product-Moment coefficient.