Open Journal of Radiology, 2012, 2, 1-9
http://dx.doi.org/10.4236/ojrad.2012.21001 Published Online March 2012 (http://www.SciRP.org/journal/ojrad)
Color Fusion of Magnetic Resonance Images Improves
Intracranial Volume Measurement in Studies of Aging
Maria del C. Valdés Hernández1*, Natalie A. Royle1, Michael R. Jackson1,
Susana Muñoz Maniega1, Lars Penke2, Mark E. Bastin3, Ian J. Deary2, Joanna M. Wardlaw1
1Brain Research Imaging Centre, University of Edinburgh, Edinburgh, UK
2Department of Psychology, University of Edinburgh, Edinburgh, UK
3Department of Medical and Radiological Sciences, University of Edinburgh, Edinburgh, UK
Received December 22, 2011; revised January 24, 2012; accepted February 2, 2012
ABSTRACT
Background: Comparison of intracranial volume (ICV) measurements in different subpopulations offers insight into
age-related atrophic change and pathological loss of neuronal tissue. For such comparisons to be meaningful the accu-
racy of ICV measurement is paramount. Color magnetic resonance images (MRI) have been utilised in several research
applications and are reported to show promise in the clinical arena. Methods: We selected a sample of 150 older com-
munity-dwelling individuals (age 71 to 72 years) representing a wide range of ICV, white matter lesions and atrophy.
We compared the extraction of ICV by thresholding on T2*-weighted MR images followed by manual editing (refer-
ence standard) done by an analyst trained in brain anatomy, with thresholding plus computational morphological opera-
tions followed by manual editing on a framework of a color fusion technique (MCMxxxVI) and two automatic brain
segmentation methods widely used, these last three done by two image analysts. Results: The range of ICV was 1074 to
1921 cm3 for the reference standard. The mean difference between the reference standard and the ICV measured using
the technique that involved the color fusion was 2.7%, while it was 5.4% compared with any fully automatic tech-
nique. However, the 95% confidence interval of the difference between the reference standard and each method was
similar: it was 7% for the segmentation aided by the color fusion and was 7% and 8.3% for the two fully automatic
methods tested. Conclusion: For studies of aging, the use of color fusion MRI in ICV segmentation in a semi-auto-
matic framework delivered best results compared with a reference standard manual method. Fully automated meth-
ods, while fast, all require manual editing to avoid significant errors and, in this post-processing step color fusion
MRI is recommended.
Keywords: Intracranial Volume; Segmentation; Brain Volume; Aging; Color MRI
1. Introduction
Brain size, assessed volumetrically, varies between peo-
ple and changes over the lifespan [1-4]. Age-related atro-
phy occurs in healthy older adults, with limited data avai-
lable to date showing that grey matter volume declines
steadily with age and white matter volume begins to de-
cline after middle-age [5]. Reduction in brain size over
the lifespan has been associated with a general decline in
cognitive performance in normal healthy older adults [6,
7].
To accurately assess the degree to which age-related
atrophy is associated with cognitive decline in healthy
aging, baseline measures of both cognitive ability and
brain size are required. In young adults the brain nor-
mally fits strongly inside the inner table of the skull,
therefore the intracranial volume (ICV) is considered
approximately the same as the maximum mature brain
size [8]. But in individuals of increasing maturity, loss of
brain tissue means that a higher proportion of ICV re-
presents cerebrospinal fluid (CSF), either within the ven-
tricles or in the extra-axial spaces. Thus ICV offers a po-
tentially reproducible measure of maximum adult brain
size in any given individual.
As methods to limit the adverse effects of aging and
brain research associated with that, are an increasing pri-
ority for many governments, validated methods of mea-
suring ICV quickly and reliably for use in large imaging
studies of aging are required. A variety of different me-
thods of calculating ICV have been suggested. However,
these methods of estimating ICV have generally not been
tested for accuracy or reliability [8]. ICV can be esti-
mated by measuring the intracranial cross-sectional area
(ICA) from a midline sagittal image, which is quicker
than full ICV measurement [9,10]. Full ICV measurement
*Corresponding author.
C
opyright © 2012 SciRes. OJRad

M. DEL C. V. HERNÁNDEZ ET AL. 2
traditionally requires manual tracing around, or thresh-
olding on, the boundary between the inner skull/CSF
interface (assuming that the dural thickness is negligible)
on each slice on which the ICV is visible. However this
is very time consuming and subject to observer bias.
Thus, some studies have used a brain volume extracted
using fully automated methods as a surrogate for ICV
[11,12], and several such techniques have been described
[13]. However, studies of automatic brain extraction
methods have shown that editing of extracranial struc-
tures is required to accurately extract brain volume [14].
This paper presents an alternative method for removing
extracranial tissue in a semi-automated way, combining
images from more than one MRI pulse sequence to pro-
duce a color image that combines attributes from each of
the original sequences.
This concept is not new. Color MR images were sug-
gested by Holland and Bottomley more than thirty years
ago [15], and the first clinical color MR images were
published in 1987 by Weiss et al. [16]. Soon after, dif-
ferent color composite techniques were implemented in
different ways to improve the information content and
enhance conspicuity of specific tissues and fluids (Table
1).
After almost a decade, in 2009, the RSNA unveiled a
new color MRI software package: Rev ColorMRI (http://
www.revolutionsmedical.com/RevMed-ColorMRI.php)
and a year later, Valdes Hernandez et al. published a
technique that fuses two different MR sequences mapped
in the red-green space to segment brain tissues and white
matter lesions: Multispectral Color Modulation and Vari-
ance Identification (MCMxxxVI) [17], and made it freely
available as an open source product (http://sourceforge.
net/projects/bric1936/). We applied this technique to ICV
segmentation and evaluated its results, comparing them
with well-established segmentation techniques widely
used in clinical research [13].
2. Materials and Methods
2.1. Subjects
We selected 150 subjects (71 - 72 years old; 87 males
and 63 females) from the Lothian Birth Cohort 1936
(LBC1936) [18]. This cohort includes 1091 relatively
healthy community dwelling older subjects born in 1936
who took part in the Scottish Mental Health Survey in
1947 (Moray House Test), a validated measure of cogni-
tive ability, thus providing early life cognitive data.
Table 1. Main features of the color MRI techniques developed after the first color images were published in 1987.
Paper Name of the technique Principles Advantages/disadvantages
Weiss K.L. et al.
1987 [16] Hybrid color MRI
Combines and displays data from two different MR
sequences of the same anatomic slice in a single
color image.
Varying spectral hues are assigned to pixel
intensities from one image.
The luminance is derived from the intensities of the
corresponding pixels of a second spatially aligned
image.
Produces images of improved information
content.
Number of luminance levels was 4. Further
work needed to increase it to 8 or 16.
Wells, M.G. et al.
1989 [25] Color composite
Combines information from two images
(T1-weighted and PD) such that the values from one
image are represented by a change in hue (color),
and the values from the other by a change in
luminance (intensity).
For the majority of cases the combined
image can effectively represent the
information normally contained in both
images-thus speeding up the viewing
process.
Kamman R.L. et al.
1989 [27] Color composite
Color images are computed from T1 and T2 using
matrix multiplication on a pixel base.
Color resolution could be modified using different
choices of the reference triangle in which the color
combinations were defined.
This method of representation offers a
means for displaying multiple features,
independent of instrumental settings.
Brown H.K. et al.
1991, 1992 and 1993
[23,24,28]
Color composite
Calculates the mean intensity values on region of
interests (ROI) and applies various colors: one to
each ROI of two spatially aligned images. The tissue
contrast patterns are optimized in the final image.
Generates seminatural-appearing color
images that possessed enhanced
conspicuity of specific tissues and fluids.
Alfano B. et al. 1992
[26] Color composite
T1 and T2 relaxation rate maps are obtained. Each
parameter is represented with one or with a
combination of two fundamental colors (red, green,
blue) into a composite quantitative color image.
Each mixture of chromatic components
represents a unique combination of the two
relaxation parameters, easily interpretable
by the human eye, without loss of anatomic
information.
Phillips W.E. et al.
1996 [22]
Multiparameter full color
composite
Based on the results of pathologic correlation and
quantitative color image analysis, applies full color
composite generation techniques to multiple MR
images: PD, T1- and T2-weighted.
Displays clinically important
neuroanatomic and neuropathologic tissues.

M. DEL C. V. HERNÁNDEZ ET AL. 3
For the present analysis, the sample was selected to re-
present the full range of early and later life cognitive abi-
lity by using the results of the Moray-House Test taken at
11 and 70 years of age [18], brain sizes, atrophy (from
none to severe) and white matter lesion load, assessed as
described in [19]. All subjects gave written informed
consent to participate.
2.2. MRI Acquisition
T2*-weighted (T2*W) and fluid attenuated inversion
recovery (FLAIR) volumes were acquired axially using a
1.5 T GE Signa Horizon HDX clinical MRI scanner
(Milwaukee, WI, USA). The T2*W imaging parameters
were TR = 940 ms; TE = 15 ms, 256 × 256 matrix size
and 80 slices with voxel size of 1 × 1 × 2 mm. FLAIR
was acquired with TR = 9002 ms; TE = 147.4 ms and TI
= 2200 ms with matrix size 256 × 256 and 40 slices
with voxel size of 1 × 1 × 4 mm with slice locations
matching every other slice of the T2*W volume. FLAIR
volume was interpolated to 80 slices of 2 mm thickness
by rigid body registration to the T2*W volume using
FMRIB’s Linear Image Registration Tool
(http://www.fmrib. ox.ac. uk/fsl/).
2.3. Image Segmentation Methods
2.3.1. General Considerations for Measuring ICV
The measurements were done by three experienced im-
age analysts all blind to each other’s measurements. Each
analyst was most familiar with the particular analysis
method they were applying and therefore the perform-
ance could be regarded as optimal for each technique.
The reference standard ICV was measured on the axial
slices of the T2*W volume, semi-automatically using
Analyze 9.0 as described in [9], by an image analyst
trained in brain anatomy.
We defined the inferior limit of the intracranial cavity
as the axial slice which was just superior to the tip of the
odontoid peg at the foramen magnum (Figure 1) and was
just inferior to the inferior limits of the cerebellar tonsils.
We excluded the cavernous sinuses/intracranial internal
carotid arteries as they enter the intracranial cavity as
these are extradural. In the reference standard ICV [9],
we included the dural venous sinuses and excluded the
contents of the sella turcica (as being outside the main
brain cavity) although we also tested the effect of in-
cluding/excluding it in the analysis because some would
argue that the sella should be excluded.
2.3.2. ICV Measurement Aided by a Color Fusion
Technique
We used the Object Extraction Tool (OET) in Analyze
9.0 which applies morphological erosion, dilation, and
region growing steps in addition to thresholding, to auto-
matically extract the contents of the intracranial cavity.
We used T2*W sequence because it offers the best dif-
ferentiation between CSF, brain tissue and inner skull
table for computational image processing.
To generate the binary mask, we placed a seed-point in
the axial slice where the orbits appear and selected the
optimal threshold as the intensity value that separated the
optic nerve from the rest of the brain tissue. In the ex-
traction process, morphological dilations were repeated
automatically to cover 99% of the voxels in the auto-
traced region on the target slice. After extraction, the
“holes” in the extracted “object” were filled in and a final
Figure 1. Midsagittal slice from a T2W MRI sequence showing how the lower limit of the intracranial volume at the foramen
magnum was defined by slice at lower limit of tonsils prior to odontoid. On the right-hand an axial view of the foramen
magnum (orange), including the slices superior (top) and inferior (bottom) to the slice defining the lower limit of the ICV
(middle).

M. DEL C. V. HERNÁNDEZ ET AL. 4
6-connected 3D region growing step was performed. The
removal of the extracranial tissues was done using the
MCMxxxVI technique (Figure 2).
The color combination facilitates the identification of
the brain boundaries in most of the controversial areas.
The combination of T2*W and FLAIR in the red/green
space used in [17] (Figure 2) does not show a good dif-
ferentiation between the dural sinuses and the CSF. If the
dural sinuses are desired to be excluded from the ICV,
the modulation of T2-weighted and T1-weighted in red
and green respectively is recommended (Figure 3 upper
row).
The color fusion principle of MCMxxxVI is the mo-
dulation of two MR sequences using two carrier signals
of wavelengths of approximately 650 nm and 510 nm
respectively. For each MR sequence, the intensity level
that represents the captured signal magnitude in each
tissue or lesion is influenced by three parameters of the
RF pulse sequence of a duration Δt used to excite the
Figure 2. MCMxxxVI. Axial slice of T2*W, FLAIR and
fused (red_green space) volumes of a scan showing the ad-
vantages of this technique differentiating the mucus con-
tained in the nasal sinuses and compensating the motion
artefact (colored image).
Figure 3. Controversial areas in which the color combina-
tion helps on the boundary definition. Upper row: trans-
versal sinuses (encircled) on T2*W (left) and fusion in the
red_green space of T2*W and FLAIR (middle) and
T2Wand T1W (right). Middle row: pituitary adenoma (en-
circled) in T2*W (left) and in the fused T2*W_FLAIR im-
age (right). Bottom row: artefact on T2*W (left) and view of
the fused T2W_T1W_T2*W on the red_green_blue color
space.
nuclear magnetic resonance signal: the tip angle (alpha),
the echo time (TE) and the pulse repetition time (TR).
Therefore, for each sequence, the intensity of each voxel
can be represented by a function I that varies in time t
(I(t)). If a sinusoidal carrier signal has the form:


0sin 2πA Atfc
t (1)



where:
A(t) is the amplitude (or magnitude) of the signal as a
function of time;
A0 is the maximum amplitude of the signal achieved in
each cycle;
fc is the frequency of oscillation (approximately 650
nm for red and 510 nm for green); and

is the phase of the signal.
When this carrier is modulated by the MR sequences,
the output signal A(t) will take the form:





00
sin 2πAA fcfIIt
tt



 

(2)
Δf delimits the range in which the output frequency
can vary respect to the carrier frequency. As Equation (2)
shows, now the carrier frequency term (in brackets) var-
ies between the extremes of fc – Δf and fc + Δf given by
the limits of variation of I(t).
The software that implements this color modulation
technique, afterwards reduces the numbers of colors by
minimum variance quantisation and graphs the resultant
colors as clusters positioned on the RGB space, thus fa-
cilitating the color discrimination of the extracranial tis-
sues and anomalies.
This method differs from the color composite tech-
niques summarised in Table 1 and from the one in which
Rev ColorMRI is based where there is an emphasis on
producing “seminatural-appearing” or “virtually realistic
appearing tissue tones”. While utilising a color scale that
mimics the “natural appearance” of the tissues may be of
benefit as an educational tool, it is unlikely that such
color scale will serve to highlight pathology. Success-
fully differentiating otherwise unseen pathology is more
likely to be achieved by the color modulation principle
explained above, and will most likely be unrelated to the
“natural” appearance of the tissues.
2.3.3. Automatic Brain Extraction
We applied two techniques: thresholding plus morpho-
logical operations using OET as explained above, but
without any manual editing to obtain a surrogate ICV;
and BET, from the FMRIB software library.
BET [13] performs a fully automatic brain extraction
in three main steps, which is used as a surrogate measure
for ICV [11,12]. Firstly it processes the intensity histo-
gram to find “robust” lower and upper intensity values
for the image, and a rough brain/non-brain threshold.
Then, it finds the centre-of-gravity of the head image,

M. DEL C. V. HERNÁNDEZ ET AL. 5
along with the rough size of the head in it. Finally, it
performs a triangular tessellation of a sphere’s surface
inside the intracranial cavity and slowly deforms it, one
vertex at a time, following forces that keep the surface
well-spaced and smooth, while attempting to move to-
wards the in- tracranial cavity’s edge in an iterative
process. This process excludes part of the lower brain
stem from the final extracted volume, although does not
guarantee that its lower limit coincides with the standard
boundary at the foramen magnum.
The BET fractional intensity threshold was optimized
by visually inspecting the brain/ICV extractions created
with a range of thresholds in a sub-sample of brain vo-
lumes. A fractional intensity threshold of 0.6 was found
to be the best compromise in our sample and subse-
quently applied to the full dataset. We did not perform
any manual editing of the derived mask obtained by this
method.
2.4. Statistical Analysis
We compared the agreement between the volumes ob-
tained by the following methods using Bland-Altman [20]
analysis of absolute and percentage differences:
1) Reference standard vs. ICV by OET followed by
MCMxxxVI, this last performed by an image analyst with
minimal knowledge of brain anatomy (usual scenario)
2) Reference standard vs. “surrogate ICV” volume by
BET
3) Reference standard vs. “surrogate ICV” volume by
OET
4) Reference standard excluding and including the pi-
tuitary gland
Pearson correlations between methods were assessed
after checking for normality using the Kolmogorov-
Smirnov test.
3. Results
The ICV, as measured by the reference standard method,
ranged from 1074 to 1921 cm3, median 1500 cm3. The
numerical values for ICV measured by each of the
methods under evaluation are summarised in Table 2.
The reference standard correlated well with all other
methods (Pearson’s r from 0.97 to 0.98, Table 2). The
effect of including or excluding the contents of the sella
turcica (mean estimated volume of this structure was
approximately 2 cm3) were minimal, with the 95% con-
fidence interval (CI) of the difference in ICV calculated
with and without the sella contents being +/– 0.49% from
the mean (7.29 cm3).
Comparison of the reference standard measurements
with the results obtained using MCMxxxVI, and the two
fully automated methods (BET and OET) (Figure 4)
showed that although the smallest mean difference was
with the automatic “surrogate ICV” extracted using OET
(2.18%), this method produced the largest variability
(95% CI +/– 8.32%, Table 2, Figure 4). The better per-
formance overall was with OET followed by MCMxxxVI
(mean difference 2.74%, 95% CI +/– 7.03%). The largest
mean difference was with ‘surrogate ICV’ by BET (mean
difference 5.38%, 95% CI +/– 6.93%).
All three automated methods produced some system-
atic bias albeit small in the measurement of ICV com-
pared with the reference standard (Figure 4). The bias
was least with automatic ICV extraction with OET and
worst with BET. This is largely attributable to inclusion
of the clivus and other extracranial structures within the
ICV which occurs particularly with BET (Figure 5).
The inclusion of the cervical spinal cord below the fo-
Table 2. Comparison of the reference standard measurements with the numeric results obtained by the methods evaluated.
Absolute intracranial volumes, correlation, differences in volumes with the 95% confidence intervals of the difference and %
differences (of mean ICV) between methods.
Methods compared Mean ICV ± SD
(cm3 )
Pearson
correlation
coefficient
(r)
Mean
difference
(cm3)*
95% CI of
difference
(cm3)
Minimum difference
(absolute value) (cm3)
Maximum difference
(absolute value) (cm3)
1502.01 ± 142.78
Reference standard vs.
ICV using OET plus editing with
MCMxxxVI 1461.73 ± 142.89 0.98 40.27
(2.74%)
105.62
(7.03%)
0.73
(0.05%)
140.58
(9.50%)
1502.01 ± 142.78
Reference standard vs.
surrogate ICV by OET 1469.86 ± 143.32 0.97 32.14 (2.18%)
62.51
(8.32%)
0.89
(0.08%)
138.80
(9.37%)
1502.01 ± 142.78
Reference standard vs.
surrogate ICV by BET 1583.68 ± 136.22 0.98 81.67
(5.38%)
104.16
(6.93%)
2.60
(0.14%)
175.22
(11.75%)
1501.03 ± 153.53
Reference standard including sella
vs.
Reference standard excluding sella 1502.01 ± 142.78 0.99 8.74
(0.29%)
7.29
(0.49%)
1.61
(0.05%)
18.06
(0.58%)
Legend: SD: standard deviation, CI: confidence interval (4SD) * % differences from the averaged measurement between the two methods being compared.

M. DEL C. V. HERNÁNDEZ ET AL. 6
Figure 4. Bland-Altman plot comparing the ICV measure-
ments done by thresholding followed by manual editing
(reference standard) with the ICV measurements done by
thresholding combined with morphological operations using
Object Extraction tool in Analyze and manual editing in
MCMxxxVI, and two different automatic techniques (Ob-
ject Extraction Tool in Analyze and Brain Extraction
Tool).
Figure 5. Sagittal and axial view of surrogate brain extrac-
tion of the same subject as Figure 3(a) using OET (a) and
BET (b). The arrows indicate erroneously included areas.
ramen magnum only caused an insignificant trend in the
difference in Bland-Altman analysis (v) (Figure 6. Posi-
tive slope of 0.0031, R2 = 0.0027).
4. Discussions
We compared a semi-automated method that uses colors
for measuring ICV with a reference standard and two
state-of-the-art automated methods. We found good agree-
ment between the ICV measurements made by the refe-
rence standard traditional thresholding method followed
by manual editing and the more automatic OET in Ana-
Figure 6. Bland-Altman showing the difference introduced
to the surrogate ICV measure due to inclusion of the spinal
cord below the foramen magnum.
lyzeTM in which morphological operations are added to
the selection of the threshold to extract the ICV fol-
lowed by minimal rectification of the ICV boundaries
using MCMxxxVI to remove structures such as orbits,
venous sinuses and other extracranial abnormalities and
helping to identify the intracranial boundaries in difficult
areas. This suggests that accurate measurement of ICV
can be obtained faster using the latter method and this
also reduces the amount of manual editing and hence
observer dependence. The pituitary gland is a structure
that causes controversy while delineating boundaries of
the ICV. We found that the difference between including
and excluding this structure was negligible and therefore
there was no significant gain from spending time ex-
cluding this structure manually. Other fully automated
methods were more variable, did not identify ICV accu-
rately, and so should not be applied without visual in-
spection and a post-processing editing step; otherwise an
error of up to 8.32% in the ICV may occur. Our results
highlight the importance of using an accurate measure of
ICV in establishing brain atrophy in healthy aging, show-
ing that the difference in methodology alone could ac-
count for 5% - 10% of the variance in a sample and
therefore could mask subtle, but valuable, age-related
changes.
The apparent contradiction of the automatic “surrogate
ICV” by OET having a larger mean value than OET plus
editing using MCMxxxVI is due to Bland Altman being
unable to account for spatial disagreement. Visual in-
spection of the OET masks show the exclusion of the
dural venous sinuses in some areas due to the automated
threshold being higher than the threshold used for the
reference standard. Thresholds are calculated using gray-
scale intensity levels. When a threshold that corresponds
to a lower intensity level is set it will include more vo-
xels in an image than the threshold corresponding to a
high intensity level. For the Bland Altman analysis we

M. DEL C. V. HERNÁNDEZ ET AL. 7
subtracted the OET, OET plus editing using MCMxxxVI
and BET volumes from the reference standard volumes
and gave the percentage difference between methods.
OET excludes intracranial structures in some areas but
includes extracranial structures below the inferior boun-
dary, when the extra step of removing the extracranial
structures is taken the volume decreases causing the per-
centage difference to increase. Further analysis using a
measure that provides information on spatial disagree-
ment between methods would clarify these issues and
highlight the importance of visually assessing segmenta-
tion output.
The strengths of our results include the use of a sample
of brain images from older subjects selected to represent
a range of cognitive and imaging factors and therefore to
be biologically relevant. We had a large sample size
which enabled us to detect quite small differences be-
tween measurement methods as well as any systematic
bias. We tested semi-automated and automated ICV
measurement methods, as well as the effect of including
or excluding specific structures from the ICV. Another
strength of our study is that while the reference standard
was produced by an analyst well-trained in brain ana-
tomy, the other techniques were applied by image analy-
sts with strong technical background and minimal know-
ledge of brain anatomy, thus reproducing the common
scenario in which some imaging research is carried out.
Therefore, our results are generalisable. A limitation,
however, is that although several approaches that com-
bine thresholding with morphological operations for au-
tomated ICV measurement have been described, we only
tested two methods of which one was programmed in a
commercial software application and the other was a
freely available academic package, both being considered
to be representative of the available state-of-art tech-
niques. We only calculated the agreement between tech-
niques. It would be advisable to apply, in addition, other
statistical measures that consider the defined similarity of
the spatial information, like the Jaccard index, although
this statistic does not inform on what differences might
be due to.
While the inclusion of non-ICV tissue, such as the
clivus and the sphenoid sinus, and the lack of having an
anatomically-defined inferior limit to the intracranial ca-
vity might not matter in younger normal people, it will
introduce increasingly large systematic errors if used to
measure ICV in older people in whom some degree of
brain atrophy is virtually universal. Our results are in-line
with previous studies that have attempted to improve
BET for brain extraction, which have suggested includ-
ing extra steps to remove non-brain structures [14]. Auto-
matic methods still require visual assessment to ensure
that extracranial structures are not included in the ICV
and that true intracranial contents are not omitted, and a
color fusion technique, like MCMxxxVI, freely available,
can be extremely useful. Image processing techniques are
improving all the time, but at present the methods tested
in this work (which are representative of most available
techniques) all require visual inspection and manual edi-
ting to derive the correct ICV and correct brain mask.
Many studies have demonstrated a variety of methods
for producing color MR images [15,16,21-27], with evi-
dence to suggest these techniques have advantages for
the reporting clinical radiologist over the traditional grey-
scale images. Over the last two decades image fusion
techniques utilising color overlay to display more than
one parameter have become commonplace in clinical ra-
diology. Examples include color Doppler flow displayed
over greyscale structural ultrasound images, nuclear me-
dicine studies such as octreoscan/PET overlaid in color
over greyscale structural CT and fusion of structural ul-
trasound images (displayed in color) with corresponding
CT or MRI cross-sectional imaging. However, as yet,
the potential for combining more than one MRI sequence
and displaying the co-registered images in color has not
been exploited within clinical radiology.
The data presented here, along with a collection of re-
lated studies, suggests color fusion MRI techniques offer
a valuable research tool. We would advocate the tech-
nique described here as a more reliable method of as-
sessing ICV, particularly in older subjects and patients
with neurological diseases with acknowledged damage
on brain imaging. We would also highlight the consider-
able benefits color fusion MR techniques could offer to
the clinical radiologist, and suggest such methods are
more thoroughly evaluated within the clinical arena. We
think it is time to give color a chance.
5. Acknowledgements
This work was funded by Age UK and the UK Medical
Research Council as part of the Study Lothian Birth Co-
hort 1936, The Centre for Cognitive Aging and Cognitive
Epidemiology (CCACE), The Row Fogo Charitable
Trust and the Scottish Founding Council through SINA-
PSE collaboration. Funding (for CCACE; G0700704/
84698) from the BBSRC, EPSRC, ESRC and MRC is
gratefully acknowledged. The imaging was performed in
the Brain Research Imaging Centre (www/sbirc.ac.uk), a
SINAPSE Centre. The DICOM to analyze image format
conversion tools used in the analysis were written by Dr.
Paul A. Armitage.
REFERENCES
[1] J. S. Allen, J. Bruss, C. K. Brown and H. Damasio, “The
Major Lobes and a Parcellation of the Temporal Region,”
Neurobiology of Aging, Vol. 26, No. 9, 2005, pp. 1245-
1260. doi:10.1016/j.neurobiolaging.2005.05.023

M. DEL C. V. HERNÁNDEZ ET AL. 8
[2] D. D. Blatter, E. D. Bigler, S. D. Gale, S. C. Johnson, C.
V. Anderson, B. M. Burnett, N. Parker, S. Kurth and S. D.
Horn, “Quantitative Volumetric Analysis of Brain MR:
Normative Database Spanning 5 Decades of Life,”
American Journal of Neuroradiology, Vol. 16, No. 2,
1995, pp. 241-251.
[3] M. A. Ikram, H. A. Vrooman, H. A. Vernooij, F. van der
Lijn, A. Hofman, A. van der Lugt, W. J. Niessen and M.
M. Breteler, “Brain Tissue Volumes in the General Eld-
erly Population: The Rotterdam Scan Study,” Neurobiol-
ogy of Aging, Vol. 29, No. 6, 2008, pp. 882-890.
doi:10.1016/j.neurobiolaging.2006.12.012
[4] R. I. Scahill, C. Frost, R. Jenkins, J. L. Whitwell, M. N.
Rosser and N. C. Fox, “A Longitudinal Study of Brain
Volume Changes in Normal Aging Using Serial Regis-
tered Magnetic Resonance Imaging,” Archives of Neu-
rology, Vol. 60, No. 7, 2003, pp. 989-994.
doi:10.1001/archneur.60.7.989
[5] Y. Ge, R. I. Grossman, J. S. Babb, M. L. Rabin and D. L.
Kolson, “Age-Related Total Gray Matter and White Mat-
ter Changes in Normal Adult Brain. Part I: Volumetric
MR Imaging Analysis,” American Journal of Neurora-
diology, Vol. 23, No. 8, 2002, pp. 1327-1333.
[6] C. E. Coffey, G. Ratcliff, J. A. Saxton, N. R. Bryan, L. P.
Fried and J. F. Lucke, “Cognitive Correlates of Human
Brain Aging: A Quantitative Magnetic Resonance Imag-
ing Investigation,” Journal of Neuropsychiatry & Clinical
Neurosciences, Vol. 13, No. 4, 2001, pp. 471-485.
doi:10.1176/appi.neuropsych.13.4.471
[7] R. T. Staff, A. D. Murray, I. J. Deary and L. J. Whalley,
“Generality and Specificity in Cognitive Aging: A Volu-
metric Brain Analysis,” Neuroimage, Vol. 30, No. 4,
2006, pp. 1433-1440.
doi:10.1016/j.neuroimage.2005.11.004
[8] G. Pengas, J. M. S. Pereira, G. B. Williams and P. J.
Nestor, “Comparative Reliability of Total Intracranial
Volume Estimation Methods and the Influence of Atro-
phy in a Longitudinal Semantic Dementia Cohort,” Jour-
nal of Neuroimaging, Vol. 19, No. 1, 2009, pp. 37-46.
doi:10.1111/j.1552-6569.2008.00246.x
[9] K. J. Ferguson, J. M. Wardlaw, C. L. Edmond, I. J. Deary,
A. M. J. McLullich and N. C. Fox, “Intracranial Area: A
Validation Method for Estimating Intracranial Volume,”
Journal of Neuroimaging, Vol. 15, No. 1, 2005, pp. 76-78.
doi:10.1111/j.1552-6569.2005.tb00289.x
[10] R. N. K. Nandigam, Y. W. Chen, M. E. Gurol, J. Rosand,
S. M. Greenberg and E. E. Smith, “Validation of Intrac-
ranial Area as a Surrogate Measure of Intracranial Vol-
ume When Using Clinical MRI,” Journal of Neuroimag-
ing, Vol. 17, No. 1, 2007, pp. 74-77.
doi:10.1111/j.1552-6569.2006.00069.x
[11] J. M. Burns, B. B. Cronk, H. S. Anderson, J. E. Donelli,
G. P. Thomas, A. Harsha, W. M. Brooks and R. H.
Swerdlow, “Cardiorespiratory Fitness and Brain Atrophy
in Early Alzheimer’s Disease,” Neurology, Vol. 71, No. 3,
2008, pp. 210-216.
doi:10.1212/01.wnl.0000317094.86209.cb
[12] R. J. Thoma, M. Monnig, F. M. Hanlon, G. A. Miller, H.
Petropoulos, A. R. Mayer, R. Yeo, M. Euler, P. Lysne, S.
N. Moses and J. M. Canive, “Hippocampus Volume and
Episodic Memory in Schizophrenia,” Journal of Interna-
tional Neuropsychological Society, Vol. 15, No. 2, 2009,
pp. 182-195. doi:10.1017/S1355617709090225
[13] S. M. Smith, “Fast Robust Automated Brain Extraction,”
Human Brain Mapping, Vol. 17, No. 3, 2010, pp. 143-
155. doi:10.1002/hbm.10062
[14] M. Battaglini, S. M. Smith, S. Brogi and N. De Stefano,
“Enhanced Brain Extraction Improves the Accuracy of
Brain Atrophy Estimation,” Neuroimage, Vol. 40, No. 2,
2008, pp. 583-589.
doi:10.1016/j.neuroimage.2007.10.067
[15] G. N. Holland and P. A. Botomley, “A Color Display
Technique for NMR Imaging,” Journal of Physics E:
Scientific Instruments, Vol. 10, No. 7, 1977, pp. 14-16.
doi:10.1088/0022-3735/10/7/014
[16] K. L. Weiss, S. O. Stiving, E. E. Herderick, J. F. Cornhill
and D. W. Chakeres, “Hybrid Color MR Imaging Dis-
play,” American Journal of Roentgenology, Vol. 149, No.
4, 1987, pp. 825-829.
[17] M. C. Valdes Hernandez, K. J. Ferguson, F. Chapell and J.
M. Wardlaw, “New Multispectral Data Fusion Technique
in MRI for White Matter Lesion Segmentation: Method
and Comparison with Thresholding in FLAIR Images,”
European Radiology, Vol. 20, No. 7, 2010, pp. 1684-
1891. doi:10.1007/s00330-010-1718-6
[18] I. J. Deary, A. J. Gow, M. D. Taylor, J. Corley, C. Brett,
V. Wilson, H. Campbell, L. J. Whalley, P. M. Visscher, D.
J. Porteous and J. M. Starr, “The Lothian Birth Cohort
1936: A Study to Examine Influences on Cognitive Aging
from Age 11 to Age 70 and Beyond,” Geriatrics, Vol. 7,
2007, pp. 28-33. doi:10.1186/1471-2318-7-28
[19] J. M. Wardlaw, M. E. Bastin, M. C. Valdes Hernandez, S.
Munoz Maniega, N. A. Royle, Z. Morris, J. Clayden, E.
Sandeman, E. Eadie, C. Murray, J. Starr and I. J. Deary,
“Brain Aging, Cognition in Youth and Old Age, and
Vascular Disease in the Lothian Birth Cohort 1936: Ra-
tionale, Design and Methodology of the Imaging Proto-
col,” International Journal of Stroke, 2011, Article in
Press.
[20] J. M. Bland and D. G. Altman, “Statistical Methods for
Assessing Agreement between Two Methods of Clinical
Measurement,” Lancet, Vol. 327, No. 8476, 1986, pp.
307-310. doi:10.1016/S0140-6736(86)90837-8
[21] M. V. Vannier, R. L. Butterfield, D. Jordan, W. A. Mur-
phy, R. G. Levit and M. Gado, “Multispectral Analysis of
Magnetic Resonance Images,” Radiology, Vol. 154, No. 1,
1985, pp. 221-224.
[22] W. E. Phillips, H. K. Brown, J. Bouza and R. E. Figueroa,
“Neuroradiologic MR Applications with Multiparametric
Color Composite Display,” Magnetic Resonance Imaging,
Vol. 14, No. 1, 1996, pp. 59-72.
doi:10.1016/0730-725X(95)02043-S
[23] H. K. Brown, T. R. Hazelton and M. L. Silbiger, “Gen-
eration of Color Composites for Enhanced Tissue Differ-
entiation in Magnetic Resonance Imaging of the Brain,”
American Journal of Anatomy, Vol. 192, No. 1, 1991, pp.
23-34. doi:10.1002/aja.1001920104
[24] H. K. Brown, T. R. Hazelton, J. V. Fiorica, A. K. Parsons,

M. DEL C. V. HERNÁNDEZ ET AL. 9
L. P. Clarke and M. L. Silbiger, “Composite and Classi-
fied Color Display in MR Imaging of the Female Pelvis,”
Magnetic Resonance Imaging, Vol. 10, No. 1, 1992, pp.
143-154. doi:10.1016/0730-725X(92)90384-C
[25] M. G. Wells, P. F. Sharp and A. N. Law, “Principles and
Appraisal of Combined Images in NMR,” Medical and
Biological Engineering and Computing, Vol. 27, No. 3,
1989, pp. 277-280. doi:10.1007/BF02441485
[26] B. Alfano, A. Brunetti, A. Ciarmiello and M. Salvatore,
“Simultaneous Display of Multiple MR Parameters with
‘Quantitative Magnetic Color Imaging’,” Journal of Com-
puter Assisted Tomography, Vol. 16, No. 4, 1992, pp.
634-640. doi:10.1097/00004728-199207000-00025
[27] R. L. Kamman, G. P. Stomp and H. J. Berendsen, “Uni-
fied Multiple-Feature Color Display for MR Images,”
Magnetic Resonance in Medicine, Vol. 9, No. 2, 1989, pp.
53-55.
[28] H. K. Brown, T. R. Hazelton, A. K. Parsons, J. V. Fiorica,
C. G. Berman and M. L. Silbiger, “PC-Based Multi-
parameter Full-Color Display for Tissue Segmentation in
MRI of Adnexal Masses,” Journal of Computer Assisted
Tomography, Vol. 17, No. 6, 1993, pp. 993-1005.
doi:10.1097/00004728-199311000-00030