This copy is for personal use only. To order printed copies, contact reprints@rsna.
org
Introduction to the Technical
Aspects of Computed Diffusion-
weighted Imaging for Radiologists
Toru Higaki, PhD
Yuko Nakamura, MD, PhD
Fuminari Tatsugami, MD, PhD
Yoko Kaichi, MD, PhD
Motonori Akagi, MD
Yuij Akiyama, RT
Yasutaka Baba, MD, PhD
Makoto Iida, MD, PhD
Kazuo Awai, MD, PhD
Abbreviations: ADC = apparent diffusion co-
efficient, CNR = contrast-to-noise ratio, DW =
diffusion-weighted, FDG = fluorodeoxyglucose,
MPG = motion probing gradient, SNR = signal-
to-noise ratio, TE = echo time
RadioGraphics 2018; 38:1131–1144
https://doi.org/10.1148/rg.2018170115
Content Codes:
From the Department of Diagnostic Radiology,
Hiroshima University, 1-2-3 Kasumi, Minami-
ku, Hiroshima 734-8551, Japan (T.H.,Y.N., F.T.,
Y.K, M.A., Y.B., M.I., K.A.); and Department of
Clinical Radiology, Hiroshima University Hospi-
tal, Hiroshima, Japan (Y.A.). Presented as an ed-
ucation exhibit at the 2016 RSNA Annual Meet-
ing. Received April 28, 2017; revision requested
September 8 and received October 6; accepted
October 17. For this journal-based SA-CME ac-
tivity, the author K.A. has provided disclosures
(see end of article); all other authors, the editor,
and the reviewers have disclosed no relevant re-
lationships. Address correspondence to T.H.
(e-mail: higaki@hiroshima-u.ac.jp).
©
RSNA, 2018
SA-CME Learning Objectives
After completing this journal-based SA-CME
activity, participants will be able to:
■■Define the principles of DW imaging.
■■Describe the basic theory and advan-
tages of computed DW imaging.
■■Identify the clinical uses and applica-
tions of computed DW imaging.
See rsna.org/learning-center-rg.
1131
Imaging Physics
Diffusion-weighted (DW) imaging is a magnetic resonance (MR)
imaging method. It is an indispensable sequence for the diagnosis
of acute cerebral infarction and is recognized as a standard tool in
oncologic imaging. Computed DW imaging refers to the synthesiz-
ing of arbitrary b-value DW images from a set of measured b-value
images by voxelwise fitting. Computed DW imaging is advanta-
geous because it generates DW images with a higher diffusion effect
than that achievable by using the MR imaging units in use today.
Additionally, computed DW imaging can reduce imaging time
while producing images characterized by a higher signal-to-noise
ratio than what the acquired DW images would display at the cor-
responding b values. By fitting input images acquired at a lower b
value and correspondingly a shorter echo time, the signal intensity
of the resulting computed DW image is closer to the ideal case.
Computed DW images are generated by employing mathematical
models that use mono-, bi-, or triexponential equations. To generate
accurate computed DW images, the appropriate model must be se-
lected, and the image parameters for the input data must be chosen
accordingly. In addition, to reduce artifacts on computed DW im-
ages, the misalignment of input data must be corrected with the aid
of image registration techniques.
©
RSNA, 2018 • radiographics.rsna.org
Introduction
Diffusion-weighted (DW) imaging is an established magnetic reso-
nance (MR) imaging method for whole-body imaging and can be
used to diagnose a wide spectrum of diseases (1–6). For example,
DW imaging is a standard tool for diagnosing ischemic stroke, as dif-
fusion changes can be observed on DW images and apparent diffu-
sion coefficient (ADC) maps earlier than with other imaging modali-
ties. Analogously, as malignant tumors tend to be characterized by
increased cellularity and typically demonstrate impeded diffusion at
imaging, the signal intensity of malignant lesions on DW images is
higher than that of the normal parenchyma (1).
A crucial parameter used in DW imaging is the b value (7),
which refers to the strength of the motion probing gradient (MPG)
pulse; the contrast ratio on DW images increases as the b value
increases. However, the absolute signal intensity of a DW image is
lower with higher b values, resulting in a decrease in the signal-to-
noise ratio (SNR). Consequently, it is generally difficult to obtain
DW images with ultra-high b values at lower magnetic fields (1).
1132 July-August 2018
Teaching Points
■■ By selecting a logarithmic scale for the ordinate axis, Equation
1 defines a straight line, which is called the monoexponential
model of diffusion. The slope of the straight line corresponds
to the ADC. From this step onward, by drawing a straight
line between the signal intensities measured at two different b
values, it is possible to compute the signal intensity that would
be observed at other arbitrary b values.
■■ The logarithmic signal intensity curves measured from actual
DW images deviate from linearity at low b values. This phe-
nomenon is related to tissue microperfusion. Microperfusion
refers to the movement of protons caused by microscopic
blood flow. This motion is macroscopically observed as an
isotropic random motion, and it is not distinguished from a
true diffusion phenomenon when a low b value is observed.
■■ By using input data from an image with a b value of 150 sec/
mm2 or higher, the effect of microperfusion is eliminated, and
computed DW images can be accurately generated using the
monoexponential model.
■■ Computed DW imaging allows for the calculation of arbitrary
b-value images by postprocessing after completion of the im-
aging examination. This benefits diagnostic ability, as a high
b value results in high lesion contrast. Additionally, increasing
the b value through computation can reduce the T2 shine-
through effect.
■■ With the computed DW imaging technique, images can be
generated at a short TE with the desired high b value, using a
short TE and low b-value images as the input data.
Computed DW images are synthetic DW
images, which are calculated from measured
DW images acquired at arbitrarily selected low b
values by voxelwise fitting (8,9). Computed DW
imaging is advantageous because it generates
DW images with a higher diffusion effect than
those obtained at MR imaging (10). In addition,
computed DW imaging can reduce imaging time
while producing images with a higher SNR than
those of the measured DW images acquired at
the corresponding b values (8,9,11–13).
In this article, we introduce the theoretical
framework of DW imaging and computed DW
imaging and the application of computed DW im-
aging in the clinical setting. We also present clinical
images obtained at computed DW imaging.
Understanding DW Imaging
The basic mechanism underlying the relationship
between the signal intensity on DW images and
the b value is expressed in Equation 1 (7–10):
Sbi bi , (1)
where bi represents the b value, Sb0 represents the
signal intensity without diffusion weighting (b =
0), Sbi represents the signal intensity at bi, and D
is the apparent diffusion coefficient (ADC). The
signal intensity on DW images decreases expo-
nentially as the b value increases.
radiographics.rsna.org
Figure 1 shows DW images with b values of 0
and 1000 sec/mm2. To calculate the ADC from
Equation 1, at least two DW images obtained at
different b values are required. Typically, an im-
age with b value equal to 0 and a DW image with
high b value are obtained and used as the input
data. The ADC can then be calculated by solving
Equation 1 for D, as shown in Equation 2:
Sbi
ln . (2)
bi
Regarding the image with the b value of 0 (Fig
1a), the contrast is similar to that of a T2-
weighted MR image, with a signal intensity
defined by Equation 3 (14):
TE
, (3)
T2
where K represents a proportionality coefficient,
p represents the proton density, and T2 represents
the spin-spin relaxation time. All of those terms
are constant characteristics of the target tissue.
Echo time (TE) in Equation 3 is determined
by the MR imaging sequence. As TE increases,
the signal intensities on DW images decrease
exponentially owing to the effect of T2 relax-
ation. As seen in Equation 3, a longer TE corre-
sponds to a lower SNR on DW images and thus
a decrease in image quality (15). Furthermore,
the superimposed T2-weighted contrast mate-
rial contaminates the “pure” diffusion signal
because the T2 relaxation progresses as the TE
is extended. This phenomenon is called the T2
shine-through effect, and it requires special atten-
tion because it may produce false-positive high-
intensity areas (ie, a high diffusion signal that is
unrelated to diffusion restriction) (1,12,16,17).
While a short TE is ideal, the duration of the
MPG pulse required to obtain a high b value
causes it to be prolonged.
Generating Computed DW Images
Monoexponential Diffusion Model
Figure 2 shows the results of Equation 1, where
Sb0 is defined as 1000 sec/mm2 and D as 2.0 ×
10−3 (teal line) and again where Sb0 is defined as
750 sec/mm2 and D as 0.5 × 10−3 (purple line).
By selecting a logarithmic scale for the or-
dinate axis, Equation 1 defines a straight line,
which is called the monoexponential model of
diffusion (1,18,19). The slope of the straight line
corresponds to the ADC. From this step onward,
by drawing a straight line between the signal
intensities measured at two different b values, it
is possible to compute the signal intensity that
RG • Volume 38 Number 4 Higaki et al 1133

Figure 1. DW images of the brain. (a) Axial

DW image of the brain demonstrates a b value
of 0 sec/mm2, without an MPG pulse. (b) Axial
DW image of the brain demonstrates a b value of
1000 sec/mm2. (c) Axial ADC map demonstrates
the calculation of a pair of DW images. Typically,
an image with a b value of 0 is obtained with the
DW image with high b value to calculate the ADC.

would be observed at other arbitrary b values.

Figure 2. Signal intensity curves of DW imaging. Equation
1 defines a straight line on the graph, Sb0 corresponds to the
intersection of the lines, and D corresponds to the slope of
the lines.
This is the fundamental theory behind computed
DW imaging (8,9,12,13,16,20).
Figure 3 shows a flow diagram illustrating
the computed DW imaging generation process.
First, the value of D is calculated from the signal
intensity measured at two b values (b0 = 0 and
600 sec/mm2) using Equation 2. Next, the signal
intensity of the computed DW image (Sbi) is
calculated using Equation 1, using as input data
the signal intensity measured at a b value of 0
(Sb0), the arbitrary target b value (bx), and the
ADC value calculated in the first step. By per-
forming these computational steps for all pixels,
computed DW images are generated. In Figure
3, an example of a computed DW image gener-
ated with the target b value of 1000 sec/mm2 is
shown as the output image.
1134 July-August 2018 radiographics.rsna.org

Figure 3. Flow chart shows the generation process of an axial computed DW image. In Step 1 (teal line), the ADC values are calcu-
lated from signal intensities measured at two b values using Equation 2. In Step 2 (purple line), the signal intensity of the computed
DW image at the arbitrary target b value (bx) is calculated using Equation 1, using the measured signal intensity of b0 and the ADC
value calculated in step 1.

Figure 4. Graph shows the signal intensity curves

of multiple tissues obtained from an abdominal DW
image. The measured signal intensity curves devi-
ate from linearity at a low b value owing to the mi-
croperfusion of the tissues.

Equation 1 can be generalized using an arbi-

trary low b value data point in place of S0 (12) to
create Equation 4:

Sbi bi , (4)

where Sb1 is the signal intensity at the b value of b1.

This means that the calculation of a computed DW
image does not require an image with a b value of 0,
and any combination of two b values can be used as
input data for the computed DW imaging process.
Biexponential Diffusion Model
The graph seen in Figure 4 shows signal inten-
sity plots, generated using the logarithmic scale
described previously, of healthy abdominal or-
gans obtained from measured DW images. The
key observation is that the logarithmic signal
intensity curves measured from actual DW im-
ages deviate from linearity at low b values. This
phenomenon is related to tissue microperfusion
(7,18,19). Microperfusion refers to the move-
ment of protons caused by microscopic blood
flow. This motion is macroscopically observed
as an isotropic random motion, and it is not
distinguished from a true diffusion phenomenon
when a low b value is observed (19).
At low b values, the effects of microperfusion
and true diffusion cannot be separated, and each
impacts the signal intensity (1,18). At a b value of
about 150–250 sec/mm2 or higher, the effect of
microperfusion is eliminated, and only true diffu-
sion contributes to the signal intensity (1,21–23).
Curves modeling the effect of both microperfu-
sion and true diffusion can be generated with a
combination of straight lines, as shown in Equa-
tion 5 and plotted in Figure 5:
Sbi 1 bi
bi , (5)
where f is the diffusion fraction component and D*
is the microperfusion-related diffusion coefficient.
RG • Volume 38 Number 4
Figure 5. Graph shows the signal intensity curve of the biex-
ponential model of diffusion. The biexponential curve is dem-
onstrated by the sum of two straight lines.
Figure 6. Graph shows the effect of the microperfusion com-
ponent when b is defined as 0 sec/mm2 and is used as input
data. If the slope of the line (ADC) is calculated from b val-
ues of 600 and 0 sec/mm2, it will be overestimated owing to
microperfusion.
When b1 is defined as 0, Equation 5 can be
rewritten to create Equation 6:
Sbi bi
bi . (6)
Equations 5 and 6 are generally referred to as
biexponential models of diffusion (24–27). The data
from at least four input images are necessary to
solve these equations and generate the computed
DW image.
With the monoexponential model, choosing a
DW image with a b value of 0 as one of the input
images may result in overestimation of the slope
of the straight line D owing to the effect of the
microperfusion component (Fig 6) (1,22,23,25).
If the computed DW image is generated under
this condition, the signal intensity of the image
will significantly deviate from the actual value.
Although this problem can be solved by gen-
erating a computed DW image using Equation 5,
this comes at the cost of a much more complicated
mathematical model and more importantly would
Higaki et al 1135
require the input data from several acquired im-
ages. However, this problem can be more easily
solved by using data from a low b value input im-
age (b ≥ 150 sec/mm2). This strategy ensures that
the relationship between the b value and the signal
intensity remains linear (1,21–23). Therefore, by
using input data from an image with a b value of
150 sec/mm2 or higher, the effect of microperfu-
sion is eliminated, and computed DW images can
be accurately generated using the monoexponen-
tial model, described in Equation 4.
Figure 7 shows a comparison between com-
puted DW images, generated from input images
with b values of 0 and 150 sec/mm2. In a com-
puted DW image generated from the b values of
a 600 and 0 sec/mm2 input data pair, the signal
intensity of the normal liver parenchyma deviates
from that which is observed on the measured DW
image. Particularly in the right lobe of the liver,
the signal intensity of the normal liver paren-
chyma appears lower than on the measured DW
image. This is the result of the overestimation of
the ADC value because the image with a b valve
of 0 sec/mm2 was used as an input image.
On the computed DW image generated from
the 600 and 150 sec/mm2 b-value input pair, the
signal intensity appears to be similar to that of
the measured DW image. It is depicted as the
low signal intensity region on the color map
inset images (Fig 7).
Triexponential Diffusion Model
The Figure 8 graph plots the signal intensity
curves of normal brain tissues measured on
actual DW images using the logarithmic scale.
Expanding the b-value range to 4000 sec/mm2
adds an additional inflection point along the in-
tensity curve. This curve should be fitted by the
combination of three straight lines, as described
in Equation 7. This is called the triexponential
model of diffusion (28–31):
Sbi fp bi Dp
ff bi Df
fs bi Ds , (7)
where fp and Dp represent the perfusion-related
diffusion fraction and coefficient, ff and Df rep-
resent the fast-free diffusion fraction and coef-
ficient, and fs and Ds represent the slow-restricted
diffusion fraction and coefficient. Equation 7 can
be simplified to Equation 8 when choosing b1
defined as 0:
Sbi fp bi Dp
ff bi D f
fs bi D s . (8)
1136 July-August 2018 radiographics.rsna.org

Figure 7. Comparison of axial computed DW images gener-

ated from b-value pairs (600 and 0 sec/mm2; 600 and 150
sec/mm2) using the monoexponential model described in
Equation 4, with corresponding axial DW color map insets.
(a) Computed DW image (cDWI) demonstrates a b value of
1000 sec/mm2, generated from input images with b defined
as 600 and 0 sec/mm2. (b) Computed DW image demon-
strates a b value of 1000 sec/mm2, generated from input im-
ages with b values of 600 and 150 sec/mm2. (c) DW image
demonstrates a b value of 1000 sec/mm2. In the computed
DW images generated from the input image pair where b is
defined as 600 and 0 sec/mm2 (a), the signal intensity of the
normal liver parenchyma (red oval in a–c) deviates from that
of the actual DW image (c). This is the result of ADC value
overestimation owing to the use of the input image where
b is defined as 0 sec/mm2. DWI = diffusion-weighted image.

When generating an ultra-high b-value computed

DW image (b > 2000 sec/mm2), the triexponen-
tial model should be used (32,33).
The triexponential model requires at least six
input images to solve the equation; however, as
is the case with the biexponential model, one can
ignore the effect of microperfusion by using an im-
age with a b value of 150 sec/mm2 or higher as the
low b-value input. As a result, Equation 7 can be
further simplified to Equation 9:
Sbi ff bi Df
fs bi Ds . (9)
Figure 8. Graph shows the signal intensity curves of brain tis-
Equation 9 is a biexponential model and can be sues obtained from a DW image. An additional inflection point
appears along the curves around b value 1500–2500 sec/mm2.
solved with data from four input images. This CSF = cerebrospinal fluid, GM = gray matter, WM = white matter.
allows for the generation of ultra-high b-value
computed DW images.
sity curve is a straight line on the semilogarithmic
Optimal b-Value Selection graph (34). In the case of the multiexponential
models, the number of b values required is depen-
b Values and Input Images.—In the case of a dent on the explicit expression of the model for-
monoexponential model requiring two input im- mula being used. In general, the SNR of the DW
ages with different b values, it is necessary to select image obtained with a low b value is higher, as the
the two points in the region where the signal inten- TE can be shortened. It is also possible to reduce
RG • Volume 38 Number 4
the influence of noise when estimating the slope
by increasing the interval between the b values.
Target b Value.—A high b value enhances the
contrast of the lesion, whereas the SNR of the
image decreases. The optimal b value is dependent
on the organ being imaged (8,9). In many cases, a
value that is slightly higher than that used during
the routine examination is selected (11,13,16,20).
Misalignment between Input Images
When generating computed DW images, it is
necessary to correct morphologic misalignment
between the input images. Because computed
DW images are calculated voxelwise from the sig-
nal intensity of matching voxel positions in each
input image, each voxel should be acquired from
the same location in the patient’s body.
Among the several causes of misalignment dur-
ing imaging, one is gross motion during the exam-
ination. Even if the patient’s body is stationary, the
position of the organ may shift during imaging ow-
ing to the patient’s heartbeat or breathing (35–37).
The influence of cardiac pulsation is apparent on
the arterial vasculature, and it has a nonnegligible
effect, especially on high b-value images, which are
sensitive to microscopic motion.
Another cause of image misalignment is the
distortion associated with magnetic field inho-
mogeneities, which can be caused by factors
such as the patient’s magnetic susceptibility and
the eddy current derived from the MPG pulses
(37,38). The MPG pulse generates eddy currents
in various metallic components of the MR imag-
ing unit, resulting in unpredictable magnetic field
inhomogeneities that can distort DW images.
Image registration algorithms provide an effi-
cient tool to correct misalignments (39–43). Image
registration is the process of transforming different
images into the same coordinate system. Available
image registration methods are rigid, affine, and
deformable registration. Such algorithms are clas-
sified on the basis of the difference in the method
applied for transforming the image (44,45). The
registration algorithms can also be classified on the
basis of the method used to calculate the degree
of alignment between images, including the sum
(mean) of squared difference (46), mutual infor-
mation (42,43), and the correlation coefficient
(40,41). Figure 9 shows computed DW images
with and without misalignment correction.
Removing Outlier Pixel Values
During the computed DW imaging generation
process, outlier pixel values may appear. Such
pixels arise from image noise or artifacts in the
input images and result in artifacts on computed
DW images. For example, in Equation 2, if the
Higaki et al 1137
pixel value of Sbi or Sb0 approaches 0 asymptoti-
cally, the ADC value asymptotically approaches
negative infinity or positive infinity. When the
ADC value is ± infinity, artifacts can be reduced
by setting the output pixel value to 0.
Clinical Advantages of Computed
DW Imaging
One of the advantages of using computed DW
imaging is that the b value can be changed arbi-
trarily. In particular, even at low magnetic fields,
at which it is difficult to acquire high b-value
DW images, such images can be generated by
computation.
In general, computed DW imaging allows for
the calculation of arbitrary b-value images by
postprocessing after completion of the imag-
ing examination (8,9). This benefits diagnostic
ability, as a high b value results in high lesion
contrast. Additionally, increasing the b value
through computation can reduce the T2 shine-
through effect. However, it is important to note
that the b value cannot be increased infinitely. As
shown in Equation 1, signal intensity decreases
as the b value increases; if the b value is set too
high, the SNR drops, resulting in considerable
image quality degradation.
Another benefit of computed DW imaging is
the use of a shorter TE. As described previously,
signal intensity decreases as the TE increases,
causing the SNR on DW images to drop. In
addition, extending TE elicits T2 shine-through
effects, potentially causing false-positive high-in-
tensity lesions to be diagnosed (16). TE elonga-
tion is associated with the long duration of the
MPG pulse required to obtain a high b-value
image. With the computed DW imaging tech-
nique, images can be generated at a short TE
with the desired high b value, using a short TE
and low b-value images as the input data.
Improving Contrast-to-Noise Ratio
There are two ways to improve the contrast-to-
noise ratio (CNR) on DW images, as described
previously. The first is to shorten the TE to
reduce image noise by inputting images with low
b values. The second way to improve CNR is to
increase the b value of the generated computed
DW image to increase the contrast between
normal tissue and the lesion (48).
Figures 10–12 show cases in which the CNR
is improved by using computed DW imaging.
Figure 10 shows a patient with acute isch-
emic stroke; Figure 11 shows brain images of
a healthy subject; and Figure 12 shows images
obtained in a patient with prostate cancer.
In the case of a patient with acute ischemic
stroke (Fig 10), it would have been desirable to
1138 July-August 2018 radiographics.rsna.org

Figure 9. Reduction of artifacts (arrowheads) in axial com-

puted DW images and ADC maps, with and without misalign-
ment correction. (a, b) Computed DW image (b = 1000 sec/
mm2) (a) and ADC map (b) generated without image registra-
tion show artifacts. (c, d) Computed DW image (b = 1000 sec/
mm2) (c) and ADC map (d) generated with image registration
show a reduction in the depiction of the artifacts. (e) DW im-
age (b =1000 sec/mm2) shows a reference standard DW image.
A b-spline based deformable image registration algorithm with
mutual information was applied using 3D Slicer (version 4.6.2,
www.slicer.org) software (47). By performing the deformable
image registration, the artifacts that appeared at the border of
the organs have been reduced.

use the triexponential model (Eq 7 or 9) to gen-

erate ultra-high b-value computed DW images,
but that was not possible because an insufficient
number of input images were obtained. Conse-
quently, on the computed DW image (Fig 10a),
the normal brain tissue appears atrophic.
Nonetheless, the area of infarction is shown
more clearly on the computed DW image than
on the measured DW image obtained on day 0.
The infarcted area shown on the computed DW
image obtained on day 0 (Fig 10a) matches that
on the measured DW image obtained on day 2
(Fig 10c), demonstrating that early lesions can
clearly be depicted on computed DW images.
In Figure 11, the measured DW image ap-
pears noisy, but this effect is reduced on the
computed DW image. Because computed DW
images are generated from low b-value images
and use a shorter TE compared with that of the
measured DW images, the noise of the computed
DW image is suppressed, improving the SNR on
the computed DW image. The signal intensity of
gray matter is higher on the computed DW image
RG • Volume 38 Number 4 Higaki et al 1139

Figure 10. Lesion (arrowhead) depiction on

axial computed DW and DW images in a 67-year-
old man with acute ischemic stroke. (a) Com-
puted DW image (cDWI) demonstrates a target b
value of 3000 sec/mm2, obtained on day 0 and
generated from input DW images with b values of
1000 and 0 sec/mm2, using the monoexponential
model. ms = msec. (b) DW image (DWI) demon-
strates a b value of 1000 sec/mm2, obtained on
day 0. (c) DW image demonstrates a b value of
1000 sec/mm2, obtained on day 2. The area of in-
farction is depicted more clearly on the computed
DW image (a) than on the actual DW image ob-
tained on day 0 (b). This observation corresponds
with the actual DW image obtained on day 2 (c).
Early lesions are clearly depicted on the computed
DW image.

Figure 11. Brain images of a healthy 34-year-old male subject. (a) Axial computed DW image shows the
brain, obtained with a target b value of 4000 sec/mm2 and a TE of 98 msec. This image was generated from
input images with b values of 250, 750, 1250, and 2000 sec/mm2, using the biexponential model described
in Equation 9. (b) Axial DW image shows the brain, obtained with a b value of 4000 sec/mm2 and a TE of
122 msec. The DW image appears noisy, but this effect is reduced on the computed DW image.
1140 July-August 2018 radiographics.rsna.org

Figure 12. Lesion depiction in a 77-year-old man with prostate cancer. (a) Axial T2-weighted MR image (T2WI) of the
prostate shows a lesion (arrowheads). (b) Axial DW image (DWI) obtained with a b value of 2000 sec/mm2 shows the
lesion (arrowheads). Dashed square = area of interest magnified in c. (c) Axial computed DW images (cDWI) with pro-
gressively increasing b value application show the lesion. The signal intensity of the lesion increases with the increasing b
values, resulting in clear lesion depiction (arrowhead) at 2000 sec/mm2. The images were generated from input images
with b values of 0 and 1000 sec/mm2, using the monoexponential model. The target b values for the computed DW im-
ages ranged from 200 to 2000 sec/mm2.

(Fig 11a) compared to that of the measured DW
image (Fig 11b) because the effect of the T2
relaxation is decreased by using a shorter TE.
In Figure 12, the signal intensity of the lesion in-
creases with the application of progressively higher
b values, resulting in clearer lesion depiction. The
diagnostic ability of computed DW imaging when
applied to prostate cancer has been reported to
improve by increasing the b value (11,13,20).
Eliminating T2 Shine-Through Effect
The computed DW imaging technique makes it
possible to generate computed DW images with
a short TE and a high b value by relying on input
images meeting these requirements. A shorter
TE and/or high b value eliminates the T2 shine-
through effect that may result in the depiction of a
false-positive high-intensity lesion. Figures 13–15
demonstrate to what extent the T2 shine-through
effect is reduced by using computed DW imaging.
Figure 13 shows images obtained in a patient with
multiple hepatic cysts; Figure 14 shows images
obtained in a patient with hepatic metastasis from
colon cancer; and Figure 15 shows images ob-
tained in a patient with brain metastasis.
In Figure 13b, the hepatic cysts appear as
high-intensity lesions. However, the ADC map
(Fig 13c) does not display diffusion restric-
tion for these lesions, which is thought to be
attributable to the T2 shine-through effect. By
generating a computed DW image with a b value
of 1500 sec/mm2, the lesions can be correctly
evaluated without referring to the ADC map.
Nakamura et al (16) reported the usefulness of
RG • Volume 38 Number 4 Higaki et al 1141

Figure 13. Hepatic cysts in a 74-year-old woman. (a) Axial computed DW image (cDWI) shows the hepatic cysts (arrowheads),
imaged with a target b value of 1500 sec/mm2, generated from input DW images with b values of 600 and 150 sec/mm2, using the
monoexponential model. ms = msec. (b) Axial DW image (DWI) shows the hepatic cysts (arrowheads) as high-intensity lesions, imaged
with a b value of 1000 sec/mm2. (c) Axial ADC map shows the hepatic cysts. By generating a computed DW image with a b value of
1500 sec/mm2, the lesions can be correctly evaluated without referring to the ADC map.

computed DW imaging for imaging noncompli-
cated hepatic cysts.
In Figure 14b, the tumor appears as a uniform
lesion on the measured DW image with a b value
of 600 sec/mm2. On the computed DW image,
only the edges of the lesion display high signal
intensity, as the signal intensity in the necrotic
region decreases with reduced T2 shine-though
effect. This matches the pattern of FDG PET
uptake (Fig 14d).
In Figure 15, on the measured DW image, the
edematous region exhibits isointense signal com-
pared with that of the surrounding region; this is
thought to be due to the T2 shine-through effect.
Since the TE is short on the computed DW im-
age, the T2 shine-through effect is reduced, and
the signal intensity of the edema decreases.
Conclusion
Computed DW imaging allows for the genera-
tion of new contrast images at postprocessing.
Although the theory underlying computed DW
imaging is simple, there are several important key
aspects that should be observed. Most notably,
when computed DW images with ultra-high b
values are generated, accurate images can be ob-
tained only by applying the triexponential model.
Even for the generation of lower b-value com-
puted DW images, the appropriate computation
models must be combined with carefully chosen
b values for the input DW images. Computed
DW imaging falls in the category of synthetic
MR imaging, which has had growing visibility
in recent years. For the widespread application
of computed DW imaging, the framework of
synthetic MR imaging may be helpful for the ef-
ficient acquisition of input images.
Disclosures of Conflicts of Interest.—K.A. Activities related
to the present article: disclosed no relevant interests. Activities
not related to the present article: research grants from Toshiba
Medical Systems, Hitachi HealthCare, Fujitsu, Eisai, Daiichi
Sankyo, Bayer Yakuhin, and Fuji Yakuhin; member of the
1142 July-August 2018 radiographics.rsna.org

Figure 14. Hepatic metastasis (arrowhead) in a 79-year-old woman. (a) Axial computed DW image (cDWI) shows the hepatic metas-
tasis with high signal intensity around the edges (b = 1500 sec/mm2). The image was generated from input DW images with b values
of 600 and 0 sec/mm2, using the monoexponential model. ms = msec. (b) Axial DW image (DWI) shows a uniform lesion, imaged with
a b value of 600 sec/mm2. (c) Axial gadolinium ethoxybenzyl–enhanced hepatobiliary phase MR image shows the hepatic metastasis.
(d) Axial fluorine 18 fluorodeoxyglucose (FDG) positron emission tomography (PET) color-mapped fusion image shows the hepatic
metastasis. The signal intensity of the tumor on the computed DW image (a) matches the pattern of FDG PET uptake.

Figure 15. T2 shine-through effect in a 75-year-old woman with brain metastasis. (a) Axial computed DW image (cDWI) of the
brain shows edema with low signal intensity, obtained with a target b value of 4000 sec/mm2. The image was generated from input
images with b values of 250, 750, 1250, and 2000 sec/mm2, using the biexponential model (Eq 9). ms = msec. (b) Axial DW image
(DWI) (b = 4000 sec/mm2) shows the edema with iso–signal intensity compared with that of the surrounding region, thought to be
due to the T2 shine-through effect. (c) Axial ADC map of the brain shows the edema. Since the TE on the computed DW image (a)
is short, the T2 shine-through effect is reduced, and the signal intensity of the edema decreases.
RG • Volume 38 Number 4
Medical Advisory Board for GE Healthcare. Other activities:
disclosed no relevant relationships.
References
1. Koh DM, Collins DJ. Diffusion-weighted MRI in the body:
applications and challenges in oncology. AJR Am J Roentgenol
2007;188(6):1622–1635.
2. Schaefer PW, Grant PE, Gonzalez RG. Diffusion-weighted
MR imaging of the brain. Radiology 2000;217(2):331–345.
3. Hagmann P, Jonasson L, Maeder P, Thiran JP, Wedeen
VJ, Meuli R. Understanding diffusion MR imaging tech-
niques: from scalar diffusion-weighted imaging to diffusion
tensor imaging and beyond. RadioGraphics 2006;26(suppl
1):S205–S223.
4. González RG, Schaefer PW, Buonanno FS, et al. Diffusion-
weighted MR imaging: diagnostic accuracy in patients
imaged within 6 hours of stroke symptom onset. Radiology
1999;210(1):155–162.
5. Srinivasan A, Goyal M, Al Azri F, Lum C. State-of-the-art
imaging of acute stroke. RadioGraphics 2006;26(suppl
1):S75–S95.
6. Rajeshkannan R, Moorthy S, Sreekumar KP, Rupa R, Prabhu
NK. Clinical applications of diffusion weighted MR imaging:
a review. Indian J Radiol Imaging 2006;16(4):705–710.
7. Le Bihan D, Breton E, Lallemand D, Grenier P, Cabanis
E, Laval-Jeantet M. MR imaging of intravoxel incoherent
motions: application to diffusion and perfusion in neurologic
disorders. Radiology 1986;161(2):401–407.
8. Blackledge MD, Leach MO, Collins DJ, Koh DM. Com-
puted diffusion-weighted MR imaging may improve tumor
detection. Radiology 2011;261(2):573–581.
9. Blackledge MD, Wilton B, Messiou C, Koh DM, Leach MO,
Collins DJ. Computed diffusion weighted imaging (cDWI)
for improving imaging contrast (abstr). In: Proceedings of the
Seventeenth Meeting of the International Society for Magnetic
Resonance in Medicine. Berkeley, Calif: International Society
for Magnetic Resonance in Medicine, 2009; 4005.
10. Kawahara S, Isoda H, Fujimoto K, et al. Additional benefit of
computed diffusion-weighted imaging for detection of hepatic
metastases at 1.5T. Clin Imaging 2016;40(3):481–485.
11. Maas MC, Fütterer JJ, Scheenen TW. Quantitative
evaluation of computed high b value diffusion-weighted
magnetic resonance imaging of the prostate. Invest Radiol
2013;48(11):779–786.
12. Gatidis S, Schmidt H, Martirosian P, Nikolaou K, Schwenzer
NF. Apparent diffusion coefficient-dependent voxelwise com-
puted diffusion-weighted imaging: an approach for improving
SNR and reducing T2 shine-through effects. J Magn Reson
Imaging 2016;43(4):824–832.
13. Yoshida R, Yoshizako T, Katsube T, Tamaki Y, Ishikawa
N, Kitagaki H. Computed diffusion-weighted imaging us-
ing 1.5-T magnetic resonance imaging for prostate cancer
diagnosis. Clin Imaging 2017;41:78–82.
14. Bammer R. Basic principles of diffusion-weighted imaging.
Eur J Radiol 2003;45(3):169–184.
15. Holdsworth SJ, Yeom KW, Antonucci MU, et al. Diffusion-
weighted imaging with dual-echo echo-planar imaging for
better sensitivity to acute stroke. AJNR Am J Neuroradiol
2014;35(7):1293–1302.
16. Nakamura Y, Higaki T, Akiyama Y, et al. Diffusion-weighted
MR imaging of non-complicated hepatic cysts: value of 3T
computed diffusion-weighted imaging. Eur J Radiol Open
2016;3:138–144.
17. Feuerlein S, Pauls S, Juchems MS, et al. Pitfalls in abdomi-
nal diffusion-weighted imaging: how predictive is restricted
water diffusion for malignancy. AJR Am J Roentgenol
2009;193(4):1070–1076.
18. Koh DM, Collins DJ, Orton MR. Intravoxel incoherent mo-
tion in body diffusion-weighted MRI: reality and challenges.
AJR Am J Roentgenol 2011;196(6):1351–1361.
19. Le Bihan D, Breton E, Lallemand D, Aubin ML, Vignaud
J, Laval-Jeantet M. Separation of diffusion and perfusion
in intravoxel incoherent motion MR imaging. Radiology
1988;168(2):497–505.
20. Ueno Y, Takahashi S, Kitajima K, et al. Computed
diffusion-weighted imaging using 3-T magnetic reso-
Higaki et al 1143
nance imaging for prostate cancer diagnosis. Eur Radiol
2013;23(12):3509–3516.
21. Barral M, Taouli B, Guiu B, et al. Diffusion-weighted MR
imaging of the pancreas: current status and recommendations.
Radiology 2015;274(1):45–63.
22. Kallehauge JF, Tanderup K, Haack S, et al. Apparent diffusion
coefficient (ADC) as a quantitative parameter in diffusion
weighted MR imaging in gynecologic cancer: dependence
on b-values used. Acta Oncol 2010;49(7):1017–1022.
23. Karki K, Hugo GD, Ford JC, et al. Estimation of optimal
b-value sets for obtaining apparent diffusion coefficient free
from perfusion in non-small cell lung cancer. Phys Med Biol
2015;60(20):7877–7891.
24. Le Bihan D, Turner R, Moonen CT, Pekar J. Imaging of
diffusion and microcirculation with gradient sensitization:
design, strategy, and significance. J Magn Reson Imaging
1991;1(1):7–28.
25. Dijkstra H, Baron P, Kappert P, Oudkerk M, Sijens PE. Ef-
fects of microperfusion in hepatic diffusion weighted imaging.
Eur Radiol 2012;22(4):891–899.
26. Chandarana H, Lee VS, Hecht E, Taouli B, Sigmund EE.
Comparison of biexponential and monoexponential model
of diffusion weighted imaging in evaluation of renal lesions:
preliminary experience. Invest Radiol 2011;46(5):285–291.
27. Lin M, Yu X, Chen Y, et al. Contribution of mono-expo-
nential, bi-exponential and stretched exponential model-
based diffusion-weighted MR imaging in the diagnosis and
differentiation of uterine cervical carcinoma. Eur Radiol
2017;27(6):2400–2410.
28. Hayashi T, Miyati T, Takahashi J, et al. Diffusion analysis
with triexponential function in liver cirrhosis. J Magn Reson
Imaging 2013;38(1):148–153.
29. Hayashi T, Miyati T, Takahashi J, et al. Diffusion analysis
with triexponential function in hepatic steatosis. Radiol Phys
Technol 2014;7(1):89–94.
30. Ueda Y, Takahashi S, Ohno N, et al. Triexponential function
analysis of diffusion-weighted MRI for diagnosing prostate
cancer. J Magn Reson Imaging 2016;43(1):138–148.
31. Cercueil JP, Petit JM, Nougaret S, et al. Intravoxel incoherent
motion diffusion-weighted imaging in the liver: comparison
of mono-, bi- and tri-exponential modelling at 3.0-T. Eur
Radiol 2015;25(6):1541–1550.
32. Hu YC, Yan LF, Sun Q, et al. Comparison between ultra-
high and conventional mono b-value DWI for preoperative
glioma grading. Oncotarget 2017;8(23):37884–37895.
33. Xueying L, Zhongping Z, Zhoushe Z, et al. Investigation
of apparent diffusion coefficient from ultra-high b-values in
Parkinson’s disease. Eur Radiol 2015;25(9):2593–2600.
34. Ogura A, Koyama D, Hayashi N, Hatano I, Osakabe K,
Yamaguchi N. Optimal b values for generation of com-
puted high-b-value DW images. AJR Am J Roentgenol
2016;206(4):713–718.
35. Pang Y, Yu B, Zhang X. Hepatic fat assessment using ad-
vanced magnetic resonance imaging. Quant Imaging Med
Surg 2012;2(3):213–218.
36. Kandpal H, Sharma R, Madhusudhan KS, Kapoor KS.
Respiratory-triggered versus breath-hold diffusion-weighted
MRI of liver lesions: comparison of image quality and ap-
parent diffusion coefficient values. AJR Am J Roentgenol
2009;192(4):915–922.
37. Chilla GS, Tan CH, Xu C, Poh CL. Diffusion weighted
magnetic resonance imaging and its recent trend: a survey.
Quant Imaging Med Surg 2015;5(3):407–422.
38. Andre JB, Bammer R. Advanced diffusion-weighted magnetic
resonance imaging techniques of the human spinal cord. Top
Magn Reson Imaging 2010;21(6):367–378.
39. Arlinghaus LR, Welch EB, Chakravarthy AB, et al. Motion
correction in diffusion-weighted MRI of the breast at 3T. J
Magn Reson Imaging 2011;33(5):1063–1070.
40. deSouza NM, Orton M, Downey K, et al. Distortion cor-
rection of echo-planar diffusion-weighted images of uterine
cervix. J Magn Reson Imaging 2016;43(5):1218–1223.
41. Hancu I, Lee SK, Hulsey K, et al. Distortion correction
in diffusion-weighted imaging of the breast: performance
assessment of prospective, retrospective, and combined
(prospective + retrospective) approaches. Magn Reson Med
2017;78(1):247–253.
1144 July-August 2018
42. Mazaheri Y, Do RK, Shukla-Dave A, Deasy JO, Lu Y, Akin
O. Motion correction of multi-b-value diffusion-weighted
imaging in the liver. Acad Radiol 2012;19(12):1573–1580.
43. Liu W, Liu X, Yang G, et al. Improving the correction of
eddy current-induced distortion in diffusion-weighted images
by excluding signals from the cerebral spinal fluid. Comput
Med Imaging Graph 2012;36(7):542–551.
44. Maintz JB, Viergever MA. A survey of medical image registra-
tion. Med Image Anal 1998;2(1):1–36.
45. Zitová B, Flusser J. Image registration methods: a survey.
Image Vis Comput 2003;21(11):977–1000.
This journal-based SA-CME activity has been approved for AMA PRA Category 1 Credit . See rsna.org/learning-center-rg.
radiographics.rsna.org
46. Higaki T, Kaneda K, Tamaki T, Date N, Azemoto S. Non-
rigid image registration for medical diagnosis using free-form
deformation with multiple grids. J Inst Image Electron Eng
Japan 2007;37(3):286–292.
47. Kapur T, Pieper S, Fedorov A, et al. Increasing the impact
of medical image computing using community-based open-
access hackathons: the NA-MIC and 3D Slicer experience.
Med Image Anal 2016;33:176–180.
48. Blackledge MD, Collins DJ, Koh DM, Leach MO. Signal
to noise ratio of high b-value diffusion weighted images is
improved using computed diffusion weighted imaging [abstr].
In: Proceedings of the Eighteenth Meeting of the International
Society for Magnetic Resonance in Medicine. Berkeley, Calif:
International Society for Magnetic Resonance in Medicine,
2010; 4707.
TM