This copy is for personal use only. To order printed copies, contact reprints@rsna.

org

Reviews and Commentary

Clinical Intravoxel Incoherent
Motion and Diffusion MR
Imaging: Past, Present, and Future1

n Review
Mami Iima, MD, PhD
The concept of diffusion magnetic resonance (MR) im-
Denis Le Bihan, MD, PhD
aging emerged in the mid-1980s, together with the first
images of water diffusion in the human brain, as a way to
probe tissue structure at a microscopic scale, although the
images were acquired at a millimetric scale. Since then,
diffusion MR imaging has become a pillar of modern clin-
ical imaging. Diffusion MR imaging has mainly been used
to investigate neurologic disorders. A dramatic applica-
tion of diffusion MR imaging has been acute brain ische-
mia, providing patients with the opportunity to receive
suitable treatment at a stage when brain tissue might still
be salvageable, thus avoiding terrible handicaps. On the
other hand, it was found that water diffusion is aniso-
tropic in white matter, because axon membranes limit
molecular movement perpendicularly to the nerve fibers.
This feature can be exploited to produce stunning maps
of the orientation in space of the white matter tracts and
brain connections in just a few minutes. Diffusion MR im-
aging is now also rapidly expanding in oncology, for the
detection of malignant lesions and metastases, as well as
monitoring. Water diffusion is usually largely decreased in
malignant tissues, and body diffusion MR imaging, which
does not require any tracer injection, is rapidly becoming
a modality of choice to detect, characterize, or even stage
malignant lesions, especially for breast or prostate cancer.
After a brief summary of the key methodological concepts
beyond diffusion MR imaging, this article will give a re-
view of the clinical literature, mainly focusing on current
outstanding issues, followed by some innovative proposals
for future improvements.

q
RSNA, 2016

Online supplemental material is available for this article.

1
From the Department of Diagnostic Imaging and Nuclear
Medicine (M.I.) and the Human Brain Research Center
(D.L.B.), Kyoto University Graduate School of Medicine,
and the Hakubi Center for Advanced Research (M.I.), Kyoto
University, Kyoto, Japan; and NeuroSpin, CEA/DSV/I2BM,
Bât 145, Point Courrier 156, CEA-Saclay Center, F-91191
Gif-sur-Yvette, France (D.L.B.). Received February 5, 2015;
revision requested April 14; final revision received May 29;
accepted June 30; final version accepted July 21. Address
correspondence to D.L.B. (e-mail: [email protected] ).

q
RSNA, 2016

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 13

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

I
n 1905 Albert Einstein published water diffusion drops immediately after After a brief summary of the key
four important articles and set the the onset of an ischemic event, when concepts beyond diffusion MR imaging,
stage for all of modern physics. One brain cells undergo swelling through we will give a review of key areas of
of his annus mirabilis articles (also his cytotoxic edema. With its unmatched clinical application. This review is of
PhD thesis dissertation) unexpectedly sensitivity, water diffusion MR imaging necessity not comprehensive, given the
gave birth to a powerful medical imag- provides patients with the opportunity huge amount of literature on the sub-
ing modality, diffusion magnetic reso- to receive suitable treatment at a stage ject. Rather, it will focus on current
nance (MR) imaging (1). The concept when brain tissue might be still salvage- outstanding issues, followed by some
of diffusion MR imaging emerged in able, thus avoiding them permanent proposals for future improvements.
the mid-1980s, together with the first loss of function. On the other hand, it
images of water molecular diffusion in was found that water diffusion is aniso-
tropic in white matter, because axon Key Concepts to Understand and
the human brain (2), as a way to probe
membranes limit molecular movement Interpret Diffusion and IVIM MR
tissue structure at a microscopic scale,
perpendicularly to the nerve fibers. Imaging
although images were acquired at a
millimetric scale (3). Since then, diffu- This feature can be exploited to pro-
sion MR imaging has become a pillar duce stunning maps of the orientation Diffusion and Brownian Motion
of modern clinical imaging. Diffusion in space of the white matter tracts and Brownian motion refers to the spon-
MR imaging is both a method and a brain connections in just a few minutes, taneous random motion of particles
powerful concept, as diffusing water as well as to provide information on suspended in a fluid. This phenome-
molecules provide unique information white matter microstructure and integ- non is named after the botanist Robert
on the tissue functional architecture. rity. With water diffusion MR imaging Brown, who in 1827 observed through
Diffusion MR imaging has mainly been it has been suggested that some psychi- a microscope that pollen grains moved
used to investigate neurologic disor- atric disorders, such as schizophrenia, through the water. Independently, the
ders. A dramatic application of diffu- might result from faulty brain connec- phenomenon of diffusion, referring to
sion MR imaging has been acute brain tions. Further work has shown (4) that the net movement of a substance from a
ischemia, following the discovery that diffusion MR imaging can work well in region of high concentration to a region
the body with free breathing and back- of low concentration, had been fully
ground signal suppression. Use of the characterized by the Fick laws. Einstein
Essentials technique is now also rapidly expand- explained later in his PhD thesis and
nn Diffusion MR imaging has ing in oncology for the detection of ma- its companion article (1) how Brownian
become a pillar of modern clin- lignant lesions and metastases, as well motion was explained by the particles
ical imaging, mainly to investi- as monitoring therapy. Water diffusion being moved by individual molecules,
gate the diseased brain, but is is substantially decreased in malignant and how their displacement was linked
also increasingly been used in tissues, and body diffusion MR imag- to the diffusion coefficient (D) of the
the body, notably in oncology. ing, which does not require any tracer Fick laws, bridging for the first time the
nn Important issues must be consid- injection, is rapidly becoming modality macroscopic diffusion and microscopic
ered when interpreting diffusion of choice to detect, characterize, or Brownian motion concepts. This expla-
MR imaging results: diffusion even stage malignant lesions, especially nation of Brownian motion served as
anisotropy, non-Gaussian diffu- breast and prostate cancer. Since its definitive confirmation that atoms and
sion, intravoxel incoherent introduction, diffusion MR imaging has molecules actually exist and was further
motion, and noise effects. enjoyed a quasi-exponential growth, verified experimentally in 1908 by Jean
with about 24 000 articles referenced in Perrin, who used it to determine the
nn Some issues still need to be PubMed in 2014 and 725 000 entries
addressed for diffusion MR im- in Google Scholar (Fig 1). However,
aging to become a clinical bio- as MR imaging scanner gradient coil Published online
marker, especially standardiza- systems, a key hardware component 10.1148/radiol.2015150244 Content code:
tion of acquisition protocols and for diffusion MR imaging, have been Radiology 2016; 278:13–32
models used for quantitative greatly improved over the recent years,
image analysis. Abbreviations:
new trends have emerged beyond the ADC = apparent diffusion coefficient
nn In the future, methods may allow original apparent diffusion coefficient D = diffusion coefficient
tissue features to be obtained (ADC) concept: IVIM is entering the D* = pseudodiffusion coefficient
directly from a limited set of dif- clinical field to evaluate tissue perfusion DTI = diffusion-tensor imaging
fusion MR imaging signals based without use of contrast agents, and the FDG = fluorodeoxyglucose
on their signature, substantially fIVIM = flowing blood volume fraction
ability to analyze non-Gaussian diffu-
IVIM = intravoxel incoherent motion
reducing acquisition and process- sion through high diffusion weighting is
ing times. boosting sensitivity to tissue features. Conflicts of interest are listed at the end of this article.

14 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

Figure 1

Figure 1: Graph shows the number of articles (vertical axis) published on diffusion and intravoxel incoherent motion
(IVIM) MR imaging since 1984, including both clinical and preclinical studies.

Avogadro number and the size of the that diffusion, although a three-dimen- and also contribute to the diffusion
water molecule (5). sional process, is only measured along signal if diffusion-coupling terms (Dxy,
With diffusion MR imaging one one direction at a time determined by Dxz) exist, which is the case when the
usually investigates the self-diffusion the orientation in space of the gradi- tissue feature axes do not coincide with
of water molecules in tissue water (dif- ent pulses. Most often diffusion is iso- the gradient directions used for mea-
fusion of other molecular moities may tropic (the same in all directions), so surements. This mean ADC is then not
also be studied with MR spectroscopy). that this spatial orientation does not rotationally invariant. Anisotropic diffu-
Diffusion-driven displacements of water matter. In some tissues, however, such sion cannot be correctly described by
molecules are encoded in the MR im- as brain white matter or muscle fibers, three diffusion coefficients along three
aging signal through variations of the diffusion is anisotropic, and diffusion directions, but requires the acquisition
magnetic field in space (6,7) caused effects strongly depend on the direction of diffusion-weighted images along at
by magnetic field gradient pulses. The of the gradient pulse. It is often thought least six different directions (diffusion-
degree of sensitivity to diffusion is de- by those in clinical practice and by MR tensor imaging [DTI] [9]). With DTI,
scribed by the so-called b value, which imaging manufacturers that one gets a one can get the trace of the diffusion
was introduced (2,8) to take into ac- “mean” diffusivity effect by averaging tensor, which represents the true mean
count the intensity and time profile of images sequentially acquired with gradi- diffusivity, indexes of the degree of an-
the gradient pulses used both for diffu- ents oriented along three perpendicular isotropy (such as fractional anisotropy),
sion encoding and MR imaging spatial directions. It can easily be shown that and so-called eigenvectors, which point
encoding. The overall effect of diffusion this is only an approximation (7), which to the directions along which diffusion
in the presence of those gradient pulses may lead to a large overestimation of is the fastest or the lowest, correspond-
is signal attenuation, and the MR imag- the true mean ADC in tissues experi- ing in general to the directions parallel
ing signal becomes diffusion weighted. encing anisotropic diffusion, especially or perpendicular to the tissue fibers,
The signal attenuation is more pro- when diffusion pulses are set on several respectively (7). DTI has served as the
nounced when large b values are used axes at the same time (to minimize echo basis for brain white matter tractog-
and when diffusion is fast. time and increase signal-to-noise ratio). raphy, but more advanced techniques
For instance, even if the diffusion-en- are currently used to take into account
Diffusion Anisotropy coding gradient pulses are set only to voxels with multiple fiber orientations
A first important consequence of the the x-axis, any gradient pulse present (6). DTI must be used in tissues where
diffusion MR imaging encoding process on the y- or z-axis will combine with the water diffusion is anisotropic, mainly
(compared with other approaches) is diffusion-encoding pulses on the x-axis in the heart, muscle, and brain white

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 15

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

matter, but other tissues may unex- if that were measureable. This concept the benefit of the method in the clinical
pectedly also show signs of anisotropy, has proved extremely powerful and du- setting. Specifically, the relationship of
such as the breast or the kidney, due to rable, and the ADC is still widely used IVIM parameters (D* and flowing blood
the presence of spatially oriented ducts today (14). volume fraction [fIVIM]) with blood
(10,11). volume and blood flow estimates using
The IVIM Concept other approaches needs to be clarified.
The ADC Concept The ADC concept was also introduced Separation of perfusion from diffusion
Another important point is that the to encompass all types of incoherent requires high signal-to-noise ratios, and
Einstein equation, which has been used motion present within each image vox- there are some technical challenges to
as a basis for diffusion MR imaging, as- el (hence, the acronym IVIM), which overcome, such artifacts from other
sumes free diffusion, as can be found in could contribute to the signal attenu- bulk flow phenomena. Vascular and
a glass of water. With free diffusion, the ation observed with diffusion MR im- tubular flow may be difficult to dis-
distribution of diffusion-driven molecu- aging, such as blood microcirculation entangle in some tissues, such as the
lar displacements obeys Gaussian law. in the capillary networks (perfusion), kidney (21). Active transport resulting
In those conditions only the diffusion and not only molecular diffusion (15). from glandular secretion (breast ducts,
coefficient, which can be obtained by Indeed, flow of blood water in randomly salivary glands and pancreas) may also
processing diffusion-weighted images, is oriented capillaries (at voxel level) be difficult to separate from microcapil-
the true diffusion coefficient. This coef- mimics a random walk (pseudodiffu- lary perfusion. One also has to keep in
ficient, which depends on temperature sion), which results in a signal atten- mind that IVIM imaging has a differen-
(2.4% change per 1°C), is around 3.0 uation in the presence of the diffusion- tial sensitivity to vessel sizes, according
3 1023 mm2/sec at 37°C. In biologic encoding gradient pulses. The effect is to the range of b values that are used.
tissues, however, diffusion is no longer seen at very low b values only, because
free, but becomes hindered by obsta- the pseudodiffusion coefficient, D*, as- The b-Value Effect
cles such as cell membranes, fibers, or sociated with blood flow is higher than Another important feature is that diffu-
macromolecules or is confined by at- the water diffusion coefficient. For this sion compared with other parameters,
tractive centers such as electric charges reason, the ADC obtained by includ- such as T1 or T2, is a genuine physical
at the proteins or cell membrane sur- ing very-low-b-value signals is usually process occurring in tissues on its own,
faces, making diffusion MR imaging higher than when larger values are not linked to MR imaging (MR imaging
exquisitely sensitive to tissue structure used (14). On the other hand, the one is merely a means to investigate it), as
in various pathologic or physiologic order of magnitude or so difference opposed to T1 or T2, which are only
conditions. The molecular displace- between true diffusion and pseudo- defined in the MR imaging context and
ment distribution then deviates from a diffusion allows them to be separated depend heavily, for instance, on the field
Gaussian law and the diffusion effect on (15,16). The idea to use diffusion and strength and MR imaging sequences. In
the MR imaging signal is no longer ade- IVIM MR imaging to get images of per- contrast, the results of diffusion MR
quately described by the Einstein equa- fusion has been found ground-breaking imaging, such as the ADC, should be,
tion. Consequently, the diffusion coef- (17), however very controversial at the in principle, equivalent across centers
ficient derived from diffusion-weighted beginning, and it took more than 20 using different MR imaging systems or
images is no longer the true diffusion years before the concept was applied sequence parameters. Unfortunately,
coefficient, but reflects interaction of in clinical practice. Indeed, IVIM MR this is true only to some extent: Prob-
water with tissue features. The model- imaging has experienced a remarkable lems may arise, as noted above, be-
ing of such diffusion effects on the sig- revival for applications throughout the cause diffusion in tissues in not free.
nal had been investigated by pioneers body over the last few years (3) (Fig With free (Gaussian) diffusion, the
such as Stejskal, Tanner, Hazlewood, 1), especially in the field of cancer im- ADC remains the same whichever set
and others (12) well before the advent aging. A key feature of IVIM diffusion of b values are used to measure it (only
of MR imaging, but this issue remains MR imaging is that it does not involve the accuracy of the ADC estimates
a complex and hot topic of investiga- contrast agents, and it may serve as an will change with the b values, and it
tion today (13). The ADC concept was interesting alternative to perfusion MR is well known that the optimal b value
introduced along with the diffusion MR imaging in some patients with contrain- for brain tissue, for instance, is around
imaging concept to avoid those diffi- dications to contrast agents or patients 1000 sec/mm2). When diffusion is
culties in a clinical setting (2). The idea with renal failure at risk for nephro- non-Gaussian, the degree of diffusion-
was to still use the Einstein equation genic systemic fibrosis (18,19) or for related signal attenuation decreases
to simply model diffusion MR imaging gadolinium deposits in brain basal gan- when the b value increases (Fig 2),
signals (as if water diffusion was Gauss- glia (20). Still, a deeper insight into the in other words, the ADC value de-
ian), but to describe the results as an IVIM concept and a clear understanding creases when high b values are used.
ADC to emphasize that results would of the strengths and limitations of the It is, thus, mandatory to indicate which
differ from the true diffusion coefficient concept are necessary to fully garner b values have been used to acquire

16 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

Figure 2

Figure 2: Diffusion MR imaging signal attenuation. Left: The natural logarithm of the signal attenuation shows a triple
curvature. At low b values (, 200 sec/mm2), the curvature results from IVIM (blood microcirculation) effects (here the
f IVIM has been set to 10%). At very high b values, the signal reaches a “noise floor,” which produces a curvature that
needs to be removed before signal analysis. The curvature visible at high b values after noise correction (deviation
from the straight line expected for free diffusion) is produced by hindrance effects (notably from membranes), which
make diffusion non-Gaussian. The kurtosis model is one approach that allows this non-Gaussian diffusion effect to be
quantified. At lower b values the signal attenuation is nearly straight, as with Gaussian diffusion. The slope obtained by
using two b values (such as 200 and 1000 sec/mm2) is smaller than the Gaussian diffusion component of the signal
ADC0, which is obtained by removing the non-Gaussian component, for instance, using the kurtosis model (Eq [1]).
Right: Since the IVIM effect is usually small, more images are often acquired at low b values than for diffusion at high
b values. However, it may be difficult to visually qualify the goodness of the diffusion and IVIM fits using the standard
attenuation plot: ln(S) as a function of b value, as in the left plot, where S is signal intensity. An attractive alternative
would be to plot S as a function of ln(b) to visually exaggerate the contribution of IVIM effects at very low b values.

data if one wishes to make meaningful Non-Gaussian Diffusion Models mT/m in clinical systems). By using
comparisons across literature. In fact, Such non-Gaussian diffusion effects such high b values, the ADC concept
not only the b values, but the precise become visible, however, only when (also often referred to as the monoex-
timing of the gradient pulses (which set high b values are used, which is now ponential model) reaches its limitation,
the diffusion time) used for diffusion possible thanks to the progress made as it cannot give a proper account of
encoding must be provided, as different in gradient hardware. While achievable the curvature of the signal attenuation
time profiles could lead to different dif- b values in the mid-1980s were in the (in semilog coordinates) that becomes
fusion effects while sharing the same b range of 100 sec/mm2, they extended apparent at high b values (Fig 2). In-
value. This is due to the fact that water to around 1000 sec/mm2 in the 1990s, deed, some extremely valuable infor-
molecules will have more chances to in- to easily reach 3000 sec/mm2 today, mation on tissue structure can be found
teract with tissue microscopic features or even above 20 000 sec/mm2 in some in this curvature (14), and several
when long diffusion times are used than prototype gradient systems made avail- models have been suggested to empir-
when short diffusion times are used, able for the Human Connectome Pro- ically handle this non-Gaussian behav-
leading to lesser signal attenuation and, ject (23) or in preclinical systems where ior, such as the polynomial or kurtosis
thus, to a smaller ADC at long diffusion gradient amplitudes of 1000 mT/m are model (24) (also called diffusion kur-
times (22). not uncommon (compared with 30–80 tosis imaging [25]), the biexponential

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 17

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

model (26), the statistical model (27), MR imaging signal (a magnitude signal to be around 200–400 sec/mm2, but
the stretched exponential model (28), that cannot be negative), there is al- may extend to 600 sec/mm2 in the brain
and others (29,30). With such models ways some background noise signal left and is expected to vary across organs
new parameters have emerged beyond and the diffusion signal remains above and pathologic conditions. In addition,
the ADC, such as the kurtosis for dif- a threshold, the noise floor (Fig 2), in- images with b of 0 sec/mm2 cannot be
fusion kurtosis imaging, which have stead of asymptotically approaching 0, acquired, as gradient pulses used for
shown great potential to characterize thus mimicking a curvature effect (16). imaging are responsible for some (tiny)
pathologic or physiologic conditions, Such noise effects must be corrected, IVIM and diffusion effects (lowest b
although they only give empirical in- if present, to avoid over- or underes- value achievable is often around 5–10
formation on the degree of diffusion timation of the model outputs (ADC, sec/mm2 or even sometimes 50 sec/
non-Gaussianity and nothing specific kurtosis, fIVIM, etc), which is not a mm2). Hence, a very good estimation
on tissue features. These models have trivial matter. Several approaches have of the theoretical signal at b of 0 sec/
been used to evaluate cerebral infarc- been proposed to correct this noise at mm2 is required to get a meaningful
tion (31), liver fibrosis (32), and tumor high b values, either by retrieving sig- estimate of fIVIM. Furthermore, while
characterization (33,34). Other models nal values from noise-corrupted data the IVIM effects reflecting microcircula-
have been designed not just to mathe- (38–42) or by using a simple proce- tion are seen only at very low b values,
matically describe the signal decay with dure where a noise-correction factor is they are usually very small and require
b values, but to provide more insight- estimated through a phantom calibra- a very good handling of the whole signal
ful, explanatory information on the tis- tion process (16). As a matter of fact, curvature, even at very large b values
sue features, mainly in the brain, such alcanes should be preferred to build (including non-Gaussian diffusion and
as the axon diameter in white matter phantoms, as they offer a wide range noise floor correction) to give correct
(composite hindered and restricted of ADC values mimicking biologic tis- estimates of parameters (16).
model of diffusion, or CHARMED (35), sues or diseases and are less prone With those concepts in mind, we
and AxCaliber models [36]) or for the to artifacts than water or ice (43) (fat will now briefly review the clinical field
gray matter neurite distribution (neu- suppression must be turned off during of applications of diffusion and IVIM
rite orientation dispersion and density measurements, though, as the alcanes MR imaging.
imaging, or NODDI, model [37]). How- resonant frequency is close to that of
ever, those extremely refined models fat). Those noise effects may explain
require strong assumptions on the un- discrepancies among the various diffu- Review of Clinical Applications
derlying tissue structure, sophisticated sion MR imaging and IVIM parameter
modeling and analysis, and still must values in the literature. Neuroimaging
be validated across the full range of Another issue related to noise is Acute and chronic stroke.—There is no
clinical conditions. It is understandable that model parameter estimates may doubt that the main application of diffu-
that clinicians might be puzzled by this depend on the algorithms that are used sion MR imaging has been for the diag-
array of diffusion models and the va- to fit the signals with the model equa- nosis of acute cerebral infarction (44),
riety of ways to process the diffusion- tions. One may fit the model equation as well as the estimation of the time
weighted images they require. However, at once (including IVIM, non-Gaussian course of ADC change in stroke (45).
clinically relevant images can be de- diffusion, and noise effects together), The ADC decrease occurring minutes
rived from parametric maps produced for instance, by using an “Exhaustive” after the ischemic insult is linked to cell
by combining images acquired with a search algorithm [16]), while a popular swelling through cytotoxic edema, but
range of b values according to the rel- way is to split the fitting into two steps, the basic mechanisms remain unclear
evant physical models by using one’s one for diffusion and the other for IVIM (12,46,47). Diffusion MR imaging has
preferred software, either in-house or effects, to increase the robustness of it- resulted in substantial changes to the
provided by vendors. Such maps often erative fitting algorithms. This is often treatment of patients with stroke, al-
allow one, in particular, to assess lesion referred to as the biexponential ap- lowing physicians to customize thera-
heterogeneity. proach (one for IVIM and one for diffu- peutic approaches (pharmacological or
sion), but should not be confused with interventional) for individual patients
Data Analysis and Noise Effects the non-Gaussian biexponential diffu- (48), as well as monitoring patient pro-
A last important issue to consider is sion model (26). In fact, taking IVIM gress on an objective basis (in both the
the effect of noise on the output pro- and non-Gaussian diffusion altogether acute and the chronic phase [49]), to
duced by such models if one wants to one should rather think of triexponen- help predict clinical outcome (48,50–
get meaningful information. There is tial or exponential-polynomial models. 52). At high degree of diffusion weigh-
still another cause of curvature of the An outstanding issue is then to decide ing, as seen through the mean kurtosis,
signal attenuation than non-Gaussian the threshold for the b value above sensitivity to tissue features increases,
diffusion at high b values: noise. At high which IVIM effects can be considered improving the characterization of is-
b values, because of the nature of the as negligible. This value is often thought chemic tissues (53). IVIM MR imaging

18 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

also has potential for the management and resting-state functional connectiv- potential to extract more microstruc-
of cerebral infarction (54) or to assess ity MR imaging have shown some dis- tural information than the ADC (34), as
the brain’s microvasculature pulsatility crepancies in their results, and further a high-degree diffusion weighting (high
on the cardiac cycle (55). investigation is needed to establish the b values) increases the effect on the
White matter diseases and tractog- relationship between structural and signal of obstacles to free diffusion pre-
raphy.—DTI has mainly been used in functional lesions associated with con- sent in tissues, notably cell membranes.
neuroscience (6); however, it is gain- cussion (65). IVIM (fIVIM) appears correlated with
ing momentum as a clinical tool. DTI vessel density (72,81), and recent stud-
can help estimate the relationship be- Oncology ies have shown a correlation between
tween tumors and nearby white matter Diffusion MR imaging has great poten- flowing blood volume fraction and ce-
tracts for preoperative and intraopera- tial as a tool in the treatment of can- rebral blood volume derived from dy-
tive planning (56). DTI is commonly cer patients, allowing earlier detection, namic susceptibility contrast-enhanced
used to investigate white matter dis- diagnosis, staging, and monitoring of MR imaging in gliomas (82,83) or dy-
orders and has also revealed faulty disease progression or response to namic contrast-enhanced–derived pa-
brain connections linked to psychiat- therapy (66). This approach is comple- rameters in renal tumors (84), head
ric disorders, such as schizophrenia, mentary to fluorodeoxyglucose (FDG) and neck tumors (85), or breast tumors
bipolar disorder, and anxiety disorder positron emission tomography (PET), (86). In light of these promising results,
(57). Diffusion-weighted imaging and which seems to be more sensitive diffusion MR imaging has been investi-
DTI have been increasingly applied in lungs and perhaps in lymph nodes gated as a potential clinical biomarker
to the clinical investigation of demy- (67), but diffusion MR imaging, which for the assessment of new drug devel-
elinating disease, especially multiple does not use ionizing radiation and any opment, as well as for monitoring drug
sclerosis, and correlations have been tracer and affords a better spatial reso- response in clinical practice (68). To
shown between diffusion-weighted im- lution, appears promising for the man- achieve this goal, several issues remain
aging findings and clinical symptoms of agement of breast, prostate, liver, and to be addressed (87) notably about the
multiple sclerosis (58). In addition, an thyroid cancers, as well as lymphomas standardization of the acquisition pro-
ADC decrease in acute disseminated (68). Furthermore, diffusion MR im- tocols (in particular fat suppression, as
encephalomyelitis has been observed, aging gives access to tissues obscured the very low diffusion coefficient of fat
which becomes more prominent dur- by sites of physiologic FDG accumula- may mimic low ADC lesions [4]) and
ing the subacute phase (59). The ADC tion, such as in the pelvis around the the models used for data processing.
decrease in the hyperacute phase of a bladder. Differences in findings are Investigations on the relationship be-
demyelinating lesion might appear even expected with FDG PET and diffusion tween the IVIM and diffusion parame-
ahead of contrast enhancement (60). MR imaging as both approaches are ters and the underlying tissue structure
Interestingly, tract-based spatial statis- based on completely different biophysi- at microscopic level, as well as changes
tics analysis of DTI data appears robust cal mechanisms. FDG PET shows areas induced by therapy, must be pursued.
than region-of-interest–based analysis with increased glucose metabolism, Reliability and reproducibility of diffu-
to predict motor outcome in primary which can also be present in inflam- sion MR imaging results must also be
progressive multiple sclerosis (61) and mation (Fig 3). ADC values correlate assessed to facilitate monitoring disease
detect widespread white matter lesions with tumor cellularity both in humans progression or response to therapy in
with a significant fractional anisotropy (69–71) and animals (72,73), and a individual patients.
decrease in patients with neuromyeli- very low pretreatment ADC is usually Brain tumors.—The ADC has been
tis optica, which is useful for the bet- associated with aggressive malignancy found useful for the differentiation of
ter understanding of the disease (62). (74,75), while relatively high pretreat- brain tumors (88), as well as tumor
DTI also has been shown to be useful ment ADC values might predict a poor grading (89), and DTI has mainly been
to assess brain lesions after mild trau- response to therapy (76–78). Some used to characterize tumor infiltration
matic brain injury, which is associated studies have shown a correlation be- or displacement to the white matter
with cognitive and physical symptoms, tween ADC values and tumor grade in around the brain tumor (90). The po-
although there are no remarkable find- humans (79,80) and animals (73). How- tential of the ADC to serve as a sur-
ings on conventional MR or computed ever, cell density is not the only histo- rogate marker for treatment response
tomographic (CT) images (63). A re- logic indicator that sets tumor grade, efficacy has emerged (91,92), and, in
cent study has revealed that fractional and other histologic features, such as combination with IVIM, to differenti-
anisotropy values in the cerebellum and nuclear atypia, may account for the im- ate high- from low-grade glioma (93).
fusiform gyri were lower in patients perfect correlation. Necrotic or cystic Moreover, regardless of tumor grade,
with mild traumatic brain injury and tumor components, which show high lower ADC values correlated with poor
vestibular symptoms, suggesting DTI ADC, could also reduce the association prognosis in malignant astrocytomas
as a diagnostic tool for the evaluation between ADC and cell density. Whole- (94). Histogram analysis of IVIM pa-
of concussion (64). Nonetheless, DTI body diffusion kurtosis imaging has the rameters may help in differentiating

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 19

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

Figure 3

20 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

recurrent tumor from treatment effect and neck tumors have been reported to seems especially promising as it allows
in glioblastoma (83). Diffusion-weight- have a higher IVIM perfusion fraction one to differentiate normal pancre-
ed imaging and DTI have the potential and ADC compared with metastatic atic parenchyma, pancreatic neoplasm
to help determine the optimal radiation lymph nodes, which could be useful in (107,108), and mass-foaming pan-
treatment volumes (95). the optimization of individualized treat- creatitis (109) with higher diagnostic
Head and neck.—The evaluation of ment planning (102). Indeed, as non- accuracy than do diffusion-weighted
the head and neck region with MR im- surgical therapeutic approaches (radi- imaging and ADC alone. IVIM MR im-
aging is hampered by susceptibility ar- ation therapy and/or chemotherapy) aging may also help identify pancreatic
tifacts, because of the contiguity of the are increasingly used in clinical practice adenocarcinoma from other pancre-
soft-tissue components with air-filled for head and neck cancer, it becomes atic masses (neuroendocrine tumor
structures and bone. In addition, some imperative to identify patients who fail and chronic pancreatitis) (110) and
specific movements (eg, jaw move- to respond to therapy as early as pos- differentiate pancreatic adenocarci-
ments, swallowing, speaking, coughing, sible after treatment has been started, nomas from neuroendocrine tumors or
or breathing), as well as respiration, so one can change or adjust the treat- benign intraductal papillary mucinous
often result in severe motion artifacts. ment regimen if necessary. Although neoplasm (111).
Some methods have been proposed to the ADC has been found to increase Prostate.—There has been growing
decrease motion artifacts (96) and to over the time-course of chemotherapy, interest in multiparametric MR imag-
overcome distortion artifacts, such as especially in squamous cell carcinoma ing including diffusion MR imaging in
read-out segmented echo-planar imag- (103,104), its potential to serve as a the detection, staging, and treatment
ing (97). Many studies have confirmed biomarker of treatment efficacy at an of prostate cancer (112). Non-Gauss-
a significant difference in ADC between early stage remains controversial and ian diffusion has already been inves-
benign and malignant lesions in this still needs validation. The differentia- tigated in the prostate with the kur-
region, but ADC values often overlap tion of posttherapeutic changes from tosis (113,114), biexponential (115),
between benign and malignant lesions. tumor recurrence is also an important or Gamma distribution (116) models.
Parameters derived from non-Gaussian clinical issue, and a recent study us- Kurtosis has a better diagnostic ability
diffusion MR imaging and IVIM might ing the IVIM model has identified both than simple ADC in differentiating
mitigate this limitation for the primary IVIM and ADC thresholds below which healthy and cancerous peripheral pros-
and nonmetastatic head and neck tu- tumor recurrence was likely (105). tate tissues (114) or low- and high-
mors (33). Diffusion and IVIM MR Pancreas.—Improvement in gra- grade prostate cancer (113) (Fig 4).
imaging have been applied to salivary dient hardware and radiofrequency Results have been mixed regarding the
gland lesions for the differentiation of coil systems allowing a good signal-to- diagnostic utility of IVIM in the diag-
benign and malignant tumor, as well noise ratio to be preserved within the nosis of prostate cancer (117,118),
as squamous cell carcinomas and lym- images while using high b values has although perfusion-free diffusion coef-
phomas (98–100). The combination of made diffusion MR imaging available ficient might have a better diagnostic
IVIM and diffusion parameters, each for the detection and characterization ability than ADC (119). In fact, IVIM
with own threshold for malignancy, re- of deep abdominal organs, such as perfusion fractions in cancer and nor-
sults in a better diagnostic ability (98). the pancreas. However, there are still mal tissue, as well as their differences,
The detection of lymph nodal metas- some challenges in using ADC values are highly variable within the literature
tases, an important factor for treatment to differentiate pancreatic cancer from (114,117,118,120–124) (Table E1 [on-
planning, (ie, defining the radiation mass-forming pancreatitis, due to the line]), and the application of IVIM in
field or the surgical neck dissection), variable proportions of fibrosis and the diagnosis of prostate cancer still
remains challenging. ADC in malignant inflammation in mass-foaming pancre- needs further validation. Diffusion MR
nodes seems substantially lower than atitis, fibrosis, necrosis, and cell den- imaging has also great potential in the
in benign nodes (101). Primary head sity in tumors (106). IVIM MR imaging active surveillance of low-risk patients,

Figure 3: Comparison of diffusion-weighted (b value = 900 sec/mm2 [b900] ) whole-body imaging with background body signal suppression (DWIBS) and FDG PET in
a 62-year-old man with small cell lung cancer showing neuroendocrine differentiation. (a) Lateral and (b) frontal maximum intensity projection (MIP) images of DWIBS and
FDG PET (inverted MIP) performed on consecutive days. Grossly there is good concordance of high uptake on FDG PET scan with high-signal-intensity (b = 900 sec/mm2)
MIP images but there are differences on (c–e) fused PET/CT (left column) and fused DWIBS and T2-weighted (right column) images. (c) Images in the upper chest. The
FDG-avid right paratracheal lymph node is not seen as a high-signal-intensity lesion on fused DWIBS/T2-weighted image (third row, arrow). Also, normal left axillary and
right internal mammary nodes (fifth row, arrows) visible on fused DWIBS/T2-weighted images are not FDG avid; these lymph nodes are likely to be normal. (d) Images in the
lower chest and liver. Normal FDG uptake in the heart not seen on fused DWIBS/T2-weighted images (third row, arrow). Liver metastases are more clearly outlined on fused
DWIBS/T2-weighted images (fourth row, arrow). Bone deposits are also more clearly seen on fused DWIBS/T2-weighted images (third row, arrow). (e) Images in the pelvis.
Note increased FDG uptake in the lower anal canal is not seen on fused DWIBS/T2-weighted images (fifth row, arrow); this is presumed to be inflammatory. Bone deposits
are more clearly seen on fused DWIBS/T2-weighted images (second and fourth rows, arrows). (Image courtesy of Prof Anwar R. Padhani, Paul Strickland Scanner Centre,
Mount Vernon Cancer Centre, London, England.)

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 21

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

Figure 4

Figure 4: Prostate cancer. Images in a 58-year-old man with a large prostate lesion involving both the left peripheral
and the transition zones (Gleason 4+4 tumor at biopsy). The panel shows the lesion on T2-weighted (T2WI) and diffu-
sion-weighted, b value of 2000 sec/mm2 (b2000 ) images (arrow), as well as ADC with combination of b values of 0, 1000
sec/mm2 (ADC0-1000 ), ADC0 (D), and kurtosis (K) maps. Signal plot shows the non-Gaussian diffusion curvature. (Image
courtesy of Prof Andrew Rosenkrantz and Prof Eric Sigmund, New York University School of Medicine, New York, NY.)

evaluation of treatment efficacy, and both perfusion from IVIM and (Gauss- architectural patterns are often pre-
prediction of disease recurrence. ian) diffusion components (126–128), sent in complex fibroadenoma (with
Breast.—Accurate differential di- leading to interesting results (Table E2 sclerosing adenosis or other compo-
agnosis of lesions, staging of malig- [online]). Recently, non-Gaussian diffu- nents of fibrocystic change) (133) and
nant lesions, as well as monitoring of sion has also been considered, giving complicated cysts, where flow patterns
treatment efficacy, are essential in the promising results for the diagnosis of might erroneously result in an increase
treatment of breast cancer. The poten- breast cancer (16,129,130), as diffu- of the flowing blood volume fraction.
tial of diffusion MR imaging to address sion kurtosis imaging parameters, in Microscopic diffusion MR imaging of
those questions is high, but results have particular, may reflect physiologic and specimens with a resolution down to
been sometimes inconsistent in the lit- morphologic alterations associated 40 µm might give new insights to the
erature partly due to differences in the with breast tumor tissues, although the understanding of the microstructural
study design (choice of b values and mechanisms need to be elucidated (31). complexity (134). A particularly chal-
acquisition methods, data analysis ap- Kurtosis is high in malignant lesions lenging problem for breast diffusion
proaches, differences in patient popu- compared with benign lesions and, MR imaging is the detection of non-
lation). The majority of the clinical in addition to ADC and flowing blood mass-enhancing lesions on dynamic
diffusion-weighted imaging breast stud- volume fraction, might improve diag- contrast-enhanced MR images. This ap-
ies rely on a monoexponential analysis, nostic accuracy (16,129), for instance, pearance is typical of ductal carcinoma
providing the simple ADC as the pa- combining parameter thresholds (131). in situ due to the tumor extension along
rameter analyzed. The combination of Fibroadenomas and fibrocystic changes the breast ducts (135,136). Manual de-
b values (0 and 1000 sec/mm2), which were found to have significant differ- lineation of regions of interest is very
include some non-Gaussian diffusion ence only in kurtosis (132). The flowing time consuming, and there is a need
effects, seems to yield the highest di- blood volume fraction is usually high in for more automatic segmentation algo-
agnostic ability to differentiate benign malignant lesions, but there seems to rithms for diffusion MR imaging to be
from malignant lesions at 1.5 T (125). be a large overlap with benign lesions used in this clinical situation. Diffusion
Some groups have attempted to extract (Table E2 [online]). More complex MR imaging has been evaluated in the

22 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

detection of lymphadenopathy in breast and cysts. Several studies have shown been shown useful for the discrimina-
cancer; however, there is a significant the potential of IVIM MR imaging tion of malignant and benign tumors, as
overlap of ADC values between benign for the evaluation of liver cirrhosis well as renal tumor subtypes (174).
and malignant lymph nodes (137–141), (155,156) or fibrosis (157), as well as
and diffusion MR imaging cannot yet re- for the tissue characterization of focal
place surgery and sentinel lymph node liver lesions (158). Combination of Future Prospects: Toward a Simplified,
biopsy for lymph-node staging. diffusion MR imaging and gadolinium Quantitative Approach for Clinical
Lung.—The evaluation of lung nod- ethoxybenzyl diethylenetriamine pen- Diffusion MR Imaging?
ules is frequently performed by using taacetic acid may increase sensitivity All those great applications have been a
contrast-enhanced CT and FDG PET for the detection of hepatic lesions, vibrant demonstration of the huge clin-
(142). Lung lesions have not been con- including liver metastases (159–162). ical potential of diffusion MR imaging,
sidered suitable for diffusion MR imag- Still, use of ADC values solely for as- but it is particularly surprising to real-
ing due to severe susceptibility artifact sessing hepatic lesions is likely to be ize that many of those diffusion MR im-
from air. Nonetheless, recent develop- challenging, as there is sometimes a aging “breakthroughs” have been based
ments in fast imaging methods such as considerable overlap between benign on empirical experimental evidence and
echo-planar imaging and parallel imag- and malignant lesions and normal liver have successfully moved into the clinical
ing have now made this possible. Quan- tissue (32,154,163,164). area with great success, but without a
titative ADC measurements in lung Luciani et al (155) found that cir- clear understanding of the mechanisms
cancer have been proposed to minimize rhotic livers had significantly decreased responsible for those findings. Water
the need of risky biopsies (143), and ADC and IVIM pseudodiffusion coef- diffusion is modulated by cell size (de-
functional diffusion maps have been ficients compared with healthy livers, creasing with cell swelling, as observed
suggested to serve as a biomarker for while the IVIM perfusion fraction has in stroke), cell density (falling with the
early prediction of treatment response been shown as a potential biomarker of increase membrane content), or cell/
in non–small cell lung carcinoma, with nonalcoholic steatohepatitis (165,166). membrane orientation (diffusion an-
a better performance than the con- isotropy in white matter fibers). But
ventional size criteria (Response Eval- Other Clinical Applications explanations for the observed find-
uation Criteria in Solid Tumors, or IVIM and diffusion MR imaging have ings have remained often qualitative.
RECIST) (144). Still, to this date, the also been used for a variety of other ap- To better understand the basis of the
results have been conflicting or incon- plications, each with its own challenges, observed findings in diffusion MR im-
clusive (145,146). such as organ motion for cardiac dif- aging, some physical modeling comes
Liver.—Detection and character- fusion MR imaging and DTI (167,168). in as a necessity. Most models have
ization of hepatocellular carcinoma Diffusion MR imaging has the potential focused on geometric features of tis-
and liver metastases, as well as pre- to differentiate benign from pathologic sues (eg, compartments such as the
diction of tumor response to ther- vertebral body compression fractures intra- and extracellular compartments,
apy, with MR imaging have benefited (169,170); however, large fat cells may physical obstacles such as fibers, cell
from liver-specific contrast agents reduce the negative correlation between membranes). Undeniably cellular com-
(147,148). However, diffusion MR im- tumor cellularity and ADC commonly ponents are largely responsible for
aging has been actively investigated found in most solid tumors (171). In the reduced diffusion coefficients in
as an alternative approach in patients the kidneys, IVIM parameters seem to biologic tissues compared with free
with severe renal failure (149). IVIM be more useful than diffusion param- water, and there is growing evidence
and diffusion MR imaging in the liver eters with the potential to predict the that membranes, even if they are per-
is degraded by artifacts due to cardiac extent of deterioration in renal function meable, are likely the main actor that
and respiratory motion (150) or to (172): D* in the renal cortex is signif- “hinders” the water diffusion process,
air in the adjacent stomach or colon. icantly lower in both mild and severe directly or indirectly. However, data on
Hence, work remains to be done to renal dysfunction, while ADC values the physical properties of water and on
establish guidelines for the acquisition decrease only in severe renal dysfunc- the status of water in biologic tissues
protocols (eg, free breathing or respi- tion. The perfusion fraction, fIVIM, and suggest that the biophysical mecha-
ratory gating, navigation, etc) so as the tissue diffusivity have shown better nisms of water diffusion in tissues may
to obtain good image quality and re- diagnostic performance, separately, not be limited to sole geometric fea-
producible IVIM and diffusion results than the overall ADC for the discrimi- tures (12). Beside protein-bounding, a
(150–153). Interestingly, the first IVIM nation of enhancing from nonenhancing large amount of water forms molecular
studies in the body were performed renal lesions, with a good correlation networks through hydrogen bonding,
in the liver by Yamada et al in 1999 between fIVIM and perfusion-related with properties, including diffusion,
(154). They showed the potential of parameters using gadolinium-based which may also be altered in the vicin-
IVIM MR imaging to differentiate he- contrast agents (173). Furthermore, ity of charged membranes. Given the
patocellular carcinoma, hemangioma, histogram analyses of IVIM data have important surface-to-volume ratio of

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 23

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

most cells, water networks interacting Figure 5

with cell membranes constitute an im-
portant fraction of tissue water. Hence,
any change in cell component’s shape,
size, or density, which induces large
variations in the membrane surface,
will impact the diffusion MR imaging
signal.
While it is very important to de-
velop such sophisticated models, we
should at the same time also make dif-
fusion MR imaging as simple and robust
as possible if one wants this outstanding
modality to expand further in the clinics
and become a standard, for instance, in
oncology, as it is today with acute brain
ischemia. The ADC concept has played
such a role (14), but we should now
also look at non-Gaussian diffusion and
IVIM effects, as they provide valuable
information on tissues. Unfortunately,
the accurate estimation of such non-
Gaussian diffusion and IVIM-related
parameters requires fitting the diffu-
sion-weighted MR imaging signal with
biophysical models using algorithms,
which are often prone to errors and
calculation-intensive, preventing real- Figure 5: Differential sensitivity of the overall IVIM/diffusion signal to each parameter according to b value
time analysis to be performed. Further- by using typical values shown in the table from reference 16. Low- and high-key b values can be identified
more, accurate data fitting with models to optimize sensitivity to IVIM (fIVIM and D* ) and to diffusion (virtual ADC that would be obtained when b
also requires the acquisition of multiple approaches 0 [ADC0 ] and kurtosis [K] ) effects, respectively.
images with a large range of b values,
resulting in long acquisition times. Ide-
ally, one should aim at decreasing ac- effects on the diffusion-weighted signal. found for malignant and benign lesions
quisition and processing times and at Taking, for instance, the IVIM-kur- in breast cancer (16), one can easily
developing new, standard (across man- tosis model, the signal intensity see (Fig 5, Fig E1 [online]) that the rel-
ufacturers’ platforms) approaches that S(b) can be written (ignoring noise ative contribution to the signal intensity
enable automatic classification of tissue floor effects for the sake of simplicity) of each parameter strongly depends on
types (ie, benign or malignant lesions) as (16): the b values, as expected from Equation
in real time. (1), but with a specific b value sensi-
S(b) = S(0){fIVIM · exp(2b · D*) tivity. For instance, the most sensitive
Key b Value and Synthetic ADC Concepts + (1 2 fIVIM) b values for fIVIM, D*, ADC0, and K
Previous literature has explored the · exp[2 b · ADC0 are around 400, 200, 800, and above
optimization of b values in the kidney + (b · ADC0)2 ·K/6]},(1) 3000 sec/mm2, respectively. One can
(175,176), liver (177,178), prostate also see that b values between 1600 and
(179), and the breast (180). However, where S(0) is the theoretical signal in- 2400 sec/mm2 are good both for ADC0
those key b values have been pro- tensity for b value of 0 sec/mm2, fIVIM and K. It also appears that variations
posed only to optimize the robustness is the (T1-, T2-weighted) volume frac- in ADC0 and K have a much greater
of the fitting results in the context of tion of incoherently flowing blood in impact on the overall signal intensity
the monoexponential model (Gaussian the tissue, b is the b value, D* is the than variations in fIVIM and D* (Fig 5),
diffusion) or to separate IVIM and dif- pseudodiffusion coefficient associated while at 400 sec/mm2 IVIM and non-
fusion effects. A completely different with the IVIM effect, ADC0 is the vir- Gaussian diffusion effects cancel each
approach would be to consider key b tual ADC that would be obtained when other.
values directly aimed at differential b approaches 0, and K is the kurtosis Based on this differential sensitivity,
diagnosis, taking into account alto- parameter. As an example, using typi- one may consider that some b values
gether IVIM and non-Gaussian diffusion cal values for fIVIM, ADC0, K, and D* (we call “key b values”) can be found to

24 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

Figure 6

Figure 6: Sindex. (a) Combined sensitivity of diffusion-weighted MR imaging signal to all IVIM and diffusion parameters. (b)
Using only the signal (voxel or region of interest level) acquired at the two key b values, an absolute Sindex can be derived that
gives an indication on the tissue nature (color scale, left). The Sindex reflects the proximity of the lesion signal to the signal
signature of typical tissues (eg, malignant, benign, liquid, etc). Example of Sindex maps (four sections) and three-dimensional
rendering show tumor heterogeneity in a rat brain 9-L glioma tumor model. T2-weighted section and histologic slice (CD31
stain) are shown for reference (see reference 72 for experimental details).

maximize sensitivity to IVIM and diffu- In summary, optimal differentiation of mm2 for brain tissue), one may now de-
sion parameters, hence to best distin- tissue types could be obtained from only fine a synthetic ADC (sADC), which is
guish tissues. Using the above values two b values (compared with many b calculated from those two key b values
as an example, two key b values can be values when the IVIM and diffusion pa- (eg, 200 and 1500 sec/mm2) as follows:
identified (Fig 6a): around 100–200 sec/ rameters have to be evaluated individu-
mm2 (“low” key b value, Lb) for IVIM ally), resulting in a dramatic shortening sADC = ln[S(Lb)/S(Hb)]/(Hb 2 Lb),(2)
and around 1400–1800 sec/mm2 (“high” of the acquisition time, an important
key b value, Hb) for non-Gaussian dif- concern for clinical protocols, especially where Lb is low-key b value, Hb is high-
fusion (mixing ADCo and kurtosis con- in noncooperative patients. Those two key b value, and S is signal intensity.
tributions). A b value above 3000 sec/ key b values are, of course, organ spe- This synthetic ADC will intrinsically in-
mm2 also has a strong tissue differen- cific (eg, body vs brain), but, according clude non-Gaussian diffusion and IVIM
tiation potential, but signal intensities to the existing literature, should be very effects in such a way as to maximize dif-
acquired at such high b values gener- similar for most body tissues. ferential sensitivity to tissue structure.
ally become very low with the gradient Following the ADC concept intro-
hardware found on typical commercial duced in the 1980s for Gaussian dif- “Sindex” Concept
MR imagers, and the kurtosis model is fusion, which is calculated from two b One may go even one step further, di-
known to fail above such high b values. values (0 and, for instance, 1000 sec/ rectly identifying tissue types based on

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 25

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

their diffusion MR imaging signal pat- Sindex is approximately 0–20, while facilitate the use of IVIM and diffusion
terns compared with signature signals, in very malignant tissues it is greater MR imaging in the clinical field, espe-
without the need to estimate any model than 100 (Fig 6b). For example, based cially for the monitoring (diagnosis and
parameter. Once those key b values on results from a previously published therapy assessment) of cancer lesions
have been identified, signature signals, breast study (16), the average Sindex (68,87). Of course, those are concepts
SM(b) and SB(b), can be calculated at was found to be 76 6 34 in malignant will have to be further developed and
those b values, once for all, for instance lesions and 31 6 34 in benign lesions. validated by using large patient cohorts
by using Equation (1), for typical tis- For a population of 46 patients with involving various organs and lesion
sues (eg, malignant [M], benign [B], re- head and neck tumors, the overall types.
spectively). Taking now the normalized performance (area under the receiver In conclusion, IVIM and non-Gauss-
signals, SV(b), from a voxel (or region of operating characteristic curve) of this ian diffusion MR imaging have the po-
interest) in a tissue under investigation Sindex to differentiate malignant from tential to give a semiautomatic diagno-
at the key b values, a “distance” (alge- benign tumors was 0.89 (sensitivity, sis of lesions with high accuracy without
braic distance, correlation coefficient, 89%; specificity, 84%; positive predic- using ionizing radiation and injection of
or scalar product, etc) can be calcu- tive value, 89%; and negative predictive radioisotopes or contrast agents. Once
lated between the vector made of these value, 84%) (181). some stabilization has been reached in
tissue signals and those of the signature In practice, this absolute scale Sin- acquisition protocol designs and in the
tissue signals using all key b values to dex derived from diffusion MR imag- models and methods used for data pro-
build a quantitative, absolute signature ing signals acquired at only two key b cessing, this approach has great poten-
index. In the simple case where only values can be obtained at voxel level tial not only to investigate neurologic
two key b values are used, as for the or within a manually or automatically and psychiatric disorders, but also in
synthetic ADC, such a signature index, drawn region of interest. Besides mean oncology for the diagnosis or staging
SI(V), can be defined as: Sindex statistics in regions of interest, of cancer lesions, as well as for drug
one may also generate histograms to development to evaluate response to
SI(V) = m
 ax{[dSV(Hb) assess lesion heterogeneity and moni- therapy.
– dSV(Lb)]/[dSM(Hb) tor therapy response, for instance, by
Acknowledgments: The authors thank the JSPS
– dSM(Lb)],0} using the standard deviation or more (Japan Society for the Promotion of Science)
– max {[dSV(Hb) advanced Sindex distribution descrip- (MI) and the Louis D. and Louis Jeantet Foun-
– dSV(Lb)]/[dSB(Hb) tors, such as skewness or kurtosis, or dations (DLB) for their generous support.
– dSB(Lb)],0},(3) determine a malignant charge by tak- Disclosures of Conflicts of Interest: M.I. Ac-
ing into account the lesion volume with tivities related to the present article: disclosed
where the relative “distance” quantities voxels for which Sindex is greater than no relevant relationships. Activities not related
to the present article: received Grant-in-Aid for
dSV,M,B(b) = [SV,M,B(b) – SN(b)]/SN(b) are 50. Color-encoded maps and three- JSPS Fellows, Grant-in-Aid for Young Scientists
calculated between the signal of the dimensional renderings of the lesions (B), Kyoto University Hakubi grant. Other re-
tissue under investigation, SV, the ma- based on the voxel-by-voxel Sindex can lationships: disclosed no relevant relationships.
lignant, SM, and the benign, SB, tissue also be generated for a better view of D.L.B. Activities related to the present article:
disclosed no relevant relationships. Activities
signals, and a reference neutral tissue lesion conspicuity and heterogeneity or not related to the present article: disclosed
signal, SN (taken as an intermediate be- to provide spatial guidance for biopsy no relevant relationships. Other relationships:
tween malignant and benign signals). sites (181) (Fig 6b, Movie [online]). author has a patent under consideration for
Sindex.
Because malignant lesions are charac- As opposed to current approaches
terized by high fIVIM, low ADCo, and based on full fitting of MR imaging
high kurtosis values, one can see that signals, which require iterative calcu- References
for malignant tissues, the signature in- lations using complex equations, the
1. Einstein A. Investigations on the theory
dex is greater than 0 (1 for the typical calculation of the Sindex is direct and of the Brownian movement. Mineola, NY:
malignant tissue of Fig 5), while it is straightforward (Eq [3]), making the Courier Dover Publications, 1956.
less than 0 for benign tissues (21 for processing time extremely short and
2. Le Bihan D, Breton E, Lallemand D,
the typical benign tissue of Fig 5). The compatible with real-time processing Grenier P, Cabanis E, Laval-Jeantet M.
signature index is 0 for a neutral (unde- (Sindex results being obtained while the MR imaging of intravoxel incoherent mo-
termined) tissue. patient is still lying in the MR imaging tions: application to diffusion and perfu-
The signature index can further be scanner) and providing opportunities to sion in neurologic disorders. Radiology
linearly scaled, if needed, for conve- perform additional scanning depending 1986;161(2):401–407.
nience, for instance as Sindex = (SI + on the results. In short, tissues or le- 3. Le Bihan D. Intravoxel incoherent motion
1) 3 25 + 25, which is now centered sions would be identified directly from perfusion MR imaging: a wake-up call. Ra-
at 50, so that Sindex is 75 for a typi- their IVIM and diffusion MR imaging diology 2008;249(3):748–752.
cal malignant tissue and 25 for a typical signal patterns without estimating any 4. Takahara T, Imai Y, Yamashita T, Yasuda
benign tissue. For normal tissues, the model parameter. This approach could S, Nasu S, Van Cauteren M. Diffusion

26 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

weighted whole body imaging with back- 18. High WA, Ayers RA, Cowper SE. Gadolin- 29. Hall MG, Barrick TR. From diffusion-
ground body signal suppression (DWIBS): ium is quantifiable within the tissue of pa- weighted MRI to anomalous diffusion imag-
technical improvement using free breath- tients with nephrogenic systemic fibrosis. J ing. Magn Reson Med 2008;59(3):447–455.
ing, STIR and high resolution 3D display. Am Acad Dermatol 2007;56(4):710–712.
30. Zhou XJ, Gao Q, Abdullah O, Magin RL.
Radiat Med 2004;22(4):275–282.
19. ACR Manual on Contrast Media. http:// Studies of anomalous diffusion in the hu-
5. Perrin J. Mouvement brownien et réalité www.acr.org/~/media/ACR/Documents/ man brain using fractional order calculus.
moléculaire. Ann Chim Phys 1909;18(8):5– PDF/QualitySafety/Resources/Contrast Magn Reson Med 2010;63(3):562–569.
114. %20Manual/2013_Contrast_Media.pdf.
Published 2013. Accessed May 15, 2015. 31. Grinberg F, Farrher E, Ciobanu L, Geffroy
6. Le Bihan D, Johansen-Berg H. Diffusion F, Le Bihan D, Shah NJ. Non-Gaussian diffu-
MRI at 25: exploring brain tissue structure 20. Kanda T, Ishii K, Kawaguchi H, Kitajima sion imaging for enhanced contrast of brain
and function. Neuroimage 2012;61(2):324– K, Takenaka D. High signal intensity in tissue affected by ischemic stroke. PLoS
341. the dentate nucleus and globus pallidus on One 2014;9(2):e89225.
unenhanced T1-weighted MR images: re-
7. Le Bihan D, Mangin JF, Poupon C, et al. Dif- 32. Anderson SW, Barry B, Soto J, Ozonoff
lationship with increasing cumulative dose
fusion tensor imaging: concepts and appli- A, O’Brien M, Jara H. Characterizing non-
of a gadolinium-based contrast material.
cations. J Magn Reson Imaging 2001;13(4): gaussian, high b-value diffusion in liver
Radiology 2014;270(3):834–841.
534–546. fibrosis: stretched exponential and diffu-
21. Notohamiprodjo M, Chandarana H, sional kurtosis modeling. J Magn Reson
8. Le Bihan D, Breton E. Imagerie de diffu- Mikheev A, et al. Combined intravoxel in- Imaging 2014;39(4):827–834.
sion in-vivo par résonance magnétique nu- coherent motion and diffusion tensor imag-
cléaire. C R Acad Sci (Paris) 1985;301(15): ing of renal diffusion and flow anisotropy. 33. Yuan J, Yeung DK, Mok GS, et al. Non-
1109–1112. Magn Reson Med 2015;73(4):1526–1532. Gaussian analysis of diffusion weighted
9. Basser PJ, Mattiello J, LeBihan D. MR dif- imaging in head and neck at 3T: a pilot
22. Pyatigorskaya N, Le Bihan D, Reynaud O, study in patients with nasopharyngeal car-
fusion tensor spectroscopy and imaging. Ciobanu L. Relationship between the diffu-
Biophys J 1994;66(1):259–267. cinoma. PLoS One 2014;9(1):e87024.
sion time and the diffusion MRI signal ob-
10. Eyal E, Shapiro-Feinberg M, Furman-Haran served at 17.2 Tesla in the healthy rat brain 34. Filli L, Wurnig M, Nanz D, Luechinger R,
E, et al. Parametric diffusion tensor imag- cortex. Magn Reson Med 2014;72(2):492– Kenkel D, Boss A. Whole-body diffusion
ing of the breast. Invest Radiol 2012;47(5): 500. kurtosis imaging: initial experience on non-
284–291. Gaussian diffusion in various organs. Invest
23. McNab JA, Edlow BL, Witzel T, et al. The
Radiol 2014;49(12):773–778.
11. Notohamiprodjo M, Glaser C, Herrmann Human Connectome Project and beyond:
KA, et al. Diffusion tensor imaging of the initial applications of 300 mT/m gradients. 35. Assaf Y, Basser PJ. Composite hin-
kidney with parallel imaging: initial clinical Neuroimage 2013;80:234–245. dered and restricted model of diffusion
experience. Invest Radiol 2008;43(10):677– (CHARMED) MR imaging of the human
24. Chabert S, Meca C, Le Bihan D. Relevance
685. brain. Neuroimage 2005;27(1):48–58.
of the information about the diffusion
12. Le Bihan D. The ‘wet mind’: water and func- distribution in vivo given by kurtosis in q- 36. Assaf Y, Blumenfeld-Katzir T, Yovel Y, Basser
tional neuroimaging. Phys Med Biol 2007; space imaging [abstr]. In: Proceedings of PJ. AxCaliber: a method for measuring axon
52(7):R57–R90. the Twelfth Meeting of the International diameter distribution from diffusion MRI.
Society for Magnetic Resonance in Medi- Magn Reson Med 2008;59(6):1347–1354.
13. Yablonskiy DA, Sukstanskii AL. Theoreti- cine. Berkeley, Calif: International Society
cal models of the diffusion weighted MR for Magnetic Resonance in Medicine, 2004; 37. Zhang H, Schneider T, Wheeler-Kingshott
signal. NMR Biomed 2010;23(7):661–681. 1238. CA, Alexander DC. NODDI: practical in
vivo neurite orientation dispersion and
14. Le Bihan D. Apparent diffusion coefficient 25. Jensen JH, Helpern JA, Ramani A, Lu H, density imaging of the human brain. Neu-
and beyond: what diffusion MR imaging Kaczynski K. Diffusional kurtosis imaging: roimage 2012;61(4):1000–1016.
can tell us about tissue structure. Radiol- the quantification of non-gaussian water
ogy 2013;268(2):318–322. diffusion by means of magnetic resonance 38. Gudbjartsson H, Patz S. The Rician
imaging. Magn Reson Med 2005;53(6): distribution of noisy MRI data. Magn Re-
15. Le Bihan D, Breton E, Lallemand D, Aubin son Med 1995;34(6):910–914.
1432–1440.
ML, Vignaud J, Laval-Jeantet M. Separa-
tion of diffusion and perfusion in intravoxel 26. Mulkern RV, Haker SJ, Maier SE. On high 39. Aja-Fernández S, Tristán-Vega A, Hoge WS.
incoherent motion MR imaging. Radiology b diffusion imaging in the human brain: ru- Statistical noise analysis in GRAPPA using
1988;168(2):497–505. minations and experimental insights. Magn a parametrized noncentral Chi approxima-
Reson Imaging 2009;27(8):1151–1162. tion model. Magn Reson Med 2011;65(4):
16. Iima M, Yano K, Kataoka M, et al. Quanti-
1195–1206.
tative non-Gaussian diffusion and intravox- 27. Yablonskiy DA, Bretthorst GL, Ackerman
el incoherent motion magnetic resonance JJ. Statistical model for diffusion attenuated 40. Brion V, Poupon C, Riff O, et al. Parallel
imaging: differentiation of malignant and MR signal. Magn Reson Med 2003;50(4): MRI noise correction: an extension of the
benign breast lesions. Invest Radiol 2015; 664–669. LMMSE to non central chi distributions. Med
50(4):205–211. Image Comput Comput Assist Interv 2011;
28. Bennett KM, Schmainda KM, Bennett RT,
14(Pt 2):226–233.
17. Dixon WT. Separation of diffusion and per- Rowe DB, Lu H, Hyde JS. Characterization
fusion in intravoxel incoherent motion MR of continuously distributed cortical water 41. Constantinides CD, Atalar E, McVeigh ER.
imaging: a modest proposal with tremen- diffusion rates with a stretched-exponential Signal-to-noise measurements in magni-
dous potential. Radiology 1988;168(2): model. Magn Reson Med 2003;50(4):727– tude images from NMR phased arrays.
566–567. 734. Magn Reson Med 1997;38(5):852–857.

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 27

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

42. Koay CG, Basser PJ. Analytically exact cor- focal ischemia. Stroke 2012;43(8):2252– 66. Patterson DM, Padhani AR, Collins DJ.
rection scheme for signal extraction from 2254. Technology insight: water diffusion MRI—
noisy magnitude MR signals. J Magn Reson a potential new biomarker of response
54. Federau C, Sumer S, Becce F, et al. Intra-
2006;179(2):317–322. to cancer therapy. Nat Clin Pract Oncol
voxel incoherent motion perfusion imaging
2008;5(4):220–233.
43. Tofts PS, Lloyd D, Clark CA, et al. Test in acute stroke: initial clinical experience.
liquids for quantitative MRI measurements Neuroradiology 2014;56(8):629–635. 67. Toledano-Massiah S, Luciani A, Itti E, et
of self-diffusion coefficient in vivo. Magn al. Whole-body diffusion-weighted imaging
55. Federau C, Hagmann P, Maeder P, et al.
Reson Med 2000;43(3):368–374. in Hodgkin lymphoma and diffuse large
Dependence of brain intravoxel incoherent
B-cell lymphoma. RadioGraphics 2015;
44. Moseley ME, Cohen Y, Mintorovitch J, motion perfusion parameters on the car-
35(3):747–764.
et al. Early detection of regional cerebral diac cycle. PLoS One 2013;8(8):e72856.
ischemia in cats: comparison of diffusion- 68. Padhani AR, Liu G, Koh DM, et al. Diffu-
56. Nimsky C, Ganslandt O, Hastreiter P, et
and T2-weighted MRI and spectroscopy. sion-weighted magnetic resonance imaging
al. Preoperative and intraoperative diffu-
Magn Reson Med 1990;14(2):330–346. as a cancer biomarker: consensus and rec-
sion tensor imaging-based fiber tracking in
ommendations. Neoplasia 2009;11(2):102–
45. Schlaug G, Siewert B, Benfield A, Edelman glioma surgery. Neurosurgery 2005;56(1):
125.
RR, Warach S. Time course of the ap- 130–137; discussion 138.
parent diffusion coefficient (ADC) abnor- 57. Shizukuishi T, Abe O, Aoki S. Diffusion 69. Guo Y, Cai YQ, Cai ZL, et al. Differentiation
mality in human stroke. Neurology 1997; tensor imaging analysis for psychiatric dis- of clinically benign and malignant breast
49(1):113–119. orders. Magn Reson Med Sci 2013;12(3): lesions using diffusion-weighted imaging. J
153–159. Magn Reson Imaging 2002;16(2):172–178.
46. Sotak CH. The role of diffusion tensor im-
aging in the evaluation of ischemic brain in- 58. Rovaris M, Gass A, Bammer R, et al. Dif- 70. Sugahara T, Korogi Y, Kochi M, et al.
jury - a review. NMR Biomed 2002;15(7-8): fusion MRI in multiple sclerosis. Neurology Usefulness of diffusion-weighted MRI with
561–569. 2005;65(10):1526–1532. echo-planar technique in the evaluation of
cellularity in gliomas. J Magn Reson Imag-
47. Norris DG, Niendorf T, Leibfritz D. Health 59. Balasubramanya KS, Kovoor JM, Jayaku- ing 1999;9(1):53–60.
and infarcted brain tissues studied at short mar PN, et al. Diffusion-weighted imag-
diffusion times: the origins of apparent ing and proton MR spectroscopy in the 71. Humphries PD, Sebire NJ, Siegel MJ, Ol-
restriction and the reduction in appar- characterization of acute disseminated sen ØE. Tumors in pediatric patients at
ent diffusion coefficient. NMR Biomed encephalomyelitis. Neuroradiology 2007; diffusion-weighted MR imaging: apparent
1994;7(7):304–310. 49(2):177–183. diffusion coefficient and tumor cellularity.
Radiology 2007;245(3):848–854.
48. Warach S, Boska M, Welch KM. Pitfalls 60. Balashov KE, Aung LL, Dhib-Jalbut S,
and potential of clinical diffusion-weighted Keller IA. Acute multiple sclerosis lesion: 72. Iima M, Reynaud O, Tsurugizawa T, et al.
MR imaging in acute stroke. Stroke 1997; conversion of restricted diffusion due to va- Characterization of glioma microcircula-
28(3):481–482. sogenic edema. J Neuroimaging 2011;21(2): tion and tissue features using intravoxel
202–204. incoherent motion magnetic resonance im-
49. Warach S, Dashe JF, Edelman RR. Clini-
aging in a rat brain model. Invest Radiol
cal outcome in ischemic stroke predicted 61. Bodini B, Cercignani M, Khaleeli Z, et al. 2014;49(7):485–490.
by early diffusion-weighted and perfusion Corpus callosum damage predicts disabil-
magnetic resonance imaging: a prelimi- ity progression and cognitive dysfunction 73. Fan G, Zang P, Jing F, Wu Z, Guo Q. Useful-
nary analysis. J Cereb Blood Flow Metab in primary-progressive MS after five years. ness of diffusion/perfusion-weighted MRI in
1996;16(1):53–59. Hum Brain Mapp 2013;34(5):1163–1172. rat gliomas: correlation with histopathology.
Acad Radiol 2005;12(5):640–651.
50. Lövblad KO, Baird AE, Schlaug G, et al. 62. Rueda Lopes FC, Doring T, Martins C, et
Ischemic lesion volumes in acute stroke by al. The role of demyelination in neuromy- 74. Turkbey B, Shah VP, Pang Y, et al. Is ap-
diffusion-weighted magnetic resonance im- elitis optica damage: diffusion-tensor MR parent diffusion coefficient associated with
aging correlate with clinical outcome. Ann imaging study. Radiology 2012;263(1):235– clinical risk scores for prostate cancers
Neurol 1997;42(2):164–170. 242. that are visible on 3-T MR images? Radiol-
ogy 2011;258(2):488–495.
51. González RG, Schaefer PW, Buonanno FS, 63. Shenton ME, Hamoda HM, Schneiderman
et al. Diffusion-weighted MR imaging: diag- JS, et al. A review of magnetic resonance 75. Nakajo M, Kajiya Y, Kaneko T, et al. FDG
nostic accuracy in patients imaged within 6 imaging and diffusion tensor imaging find- PET/CT and diffusion-weighted imaging for
hours of stroke symptom onset. Radiology ings in mild traumatic brain injury. Brain breast cancer: prognostic value of maxi-
1999;210(1):155–162. Imaging Behav 2012;6(2):137–192. mum standardized uptake values and ap-
parent diffusion coefficient values of the
52. Dreher W, Kühn B, Gyngell ML, et al. Tem- 64. Alhilali LM, Yaeger K, Collins M, Fakhran
primary lesion. Eur J Nucl Med Mol Imag-
poral and regional changes during focal is- S. Detection of central white matter in-
ing 2010;37(11):2011–2020.
chemia in rat brain studied by proton spec- jury underlying vestibulopathy after mild
troscopic imaging and quantitative diffusion traumatic brain injury. Radiology 2014; 76. Koh DM, Scurr E, Collins D, et al. Predict-
NMR imaging. Magn Reson Med 1998; 272(1):224–232. ing response of colorectal hepatic metas-
39(6):878–888. tasis: value of pretreatment apparent dif-
65. Chong CD, Schwedt TJ. White matter dam-
fusion coefficients. AJR Am J Roentgenol
53. Cheung JS, Wang E, Lo EH, Sun PZ. Strat- age and brain network alterations in con-
2007;188(4):1001–1008.
ification of heterogeneous diffusion MRI cussed patients: a review of recent diffusion
ischemic lesion with kurtosis imaging: eval- tensor imaging and resting-state functional 77. Cui Y, Zhang XP, Sun YS, Tang L, Shen
uation of mean diffusion and kurtosis MRI connectivity data. Curr Pain Headache L. Apparent diffusion coefficient: poten-
mismatch in an animal model of transient Rep 2015;19(5):485. tial imaging biomarker for prediction and

28 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

early detection of response to chemother- brain tumors at MR imaging. Radiology 100. Sumi M, Nakamura T. Head and neck tu-
apy in hepatic metastases. Radiology 2008; 2005;235(3):985–991. mors: assessment of perfusion-related pa-
248(3):894–900. rameters and diffusion coefficients based
89. Bulakbasi N, Guvenc I, Onguru O, Erdogan on the intravoxel incoherent motion model.
78. Sun YS, Zhang XP, Tang L, et al. Locally E, Tayfun C, Ucoz T. The added value of the AJNR Am J Neuroradiol 2013;34(2):410–
advanced rectal carcinoma treated with apparent diffusion coefficient calculation to 416.
preoperative chemotherapy and radiation magnetic resonance imaging in the differ-
therapy: preliminary analysis of diffusion- entiation and grading of malignant brain 101. Vandecaveye V, De Keyzer F, Vander Poorten
weighted MR imaging for early detection tumors. J Comput Assist Tomogr 2004; V, et al. Head and neck squamous cell car-
of tumor histopathologic downstaging. Ra- 28(6):735–746. cinoma: value of diffusion-weighted MR
diology 2010;254(1):170–178. imaging for nodal staging. Radiology 2009;
90. Assaf Y, Pasternak O. Diffusion tensor im-
79. Yabuuchi H, Soeda H, Matsuo Y, et al. 251(1):134–146.
aging (DTI)-based white matter mapping in
Phyllodes tumor of the breast: correlation brain research: a review. J Mol Neurosci 102. Lu Y, Jansen JF, Stambuk HE, et al. Com-
between MR findings and histologic grade. 2008;34(1):51–61. paring primary tumors and metastatic
Radiology 2006;241(3):702–709.
nodes in head and neck cancer using intra-
91. Chenevert TL, Stegman LD, Taylor JM, et
80. Higano S, Yun X, Kumabe T, et al. Malignant voxel incoherent motion imaging: a prelim-
al. Diffusion magnetic resonance imaging:
astrocytic tumors: clinical importance of inary experience. J Comput Assist Tomogr
an early surrogate marker of therapeutic
apparent diffusion coefficient in prediction 2013;37(3):346–352.
efficacy in brain tumors. J Natl Cancer Inst
of grade and prognosis. Radiology 2006;
2000;92(24):2029–2036. 103. Kim S, Loevner L, Quon H, et al. Diffusion-
241(3):839–846.
weighted magnetic resonance imaging for
81. Lee HJ, Rha SY, Chung YE, et al. Tumor 92. Hamstra DA, Chenevert TL, Moffat BA,
predicting and detecting early response to
perfusion-related parameter of diffusion- et al. Evaluation of the functional diffusion
chemoradiation therapy of squamous cell
weighted magnetic resonance imaging: map as an early biomarker of time-to-pro-
carcinomas of the head and neck. Clin
correlation with histological microvessel gression and overall survival in high-grade
Cancer Res 2009;15(3):986–994.
density. Magn Reson Med 2014;71(4): glioma. Proc Natl Acad Sci U S A 2005;
1554–1558. 102(46):16759–16764. 104. Vandecaveye V, Dirix P, De Keyzer F, et
al. Diffusion-weighted magnetic reso-
82. Federau C, Meuli R, O’Brien K, Maeder 93. Hu YC, Yan LF, Wu L, et al. Intravoxel
nance imaging early after chemoradio-
P, Hagmann P. Perfusion measurement incoherent motion diffusion-weighted MR
therapy to monitor treatment response in
in brain gliomas with intravoxel incoher- imaging of gliomas: efficacy in preoperative
head-and-neck squamous cell carcinoma.
ent motion MRI. AJNR Am J Neuroradiol grading. Sci Rep 2014;4:7208.
Int J Radiat Oncol Biol Phys 2012;82(3):
2014;35(2):256–262.
94. Zulfiqar M, Yousem DM, Lai H. ADC values 1098–1107.
83. Kim HS, Suh CH, Kim N, Choi CG, Kim and prognosis of malignant astrocytomas:
105. Tshering Vogel DW, Zbaeren P, Geretsch-
SJ. Histogram analysis of intravoxel inco- does lower ADC predict a worse progno-
laeger A, Vermathen P, De Keyzer F,
herent motion for differentiating recurrent sis independent of grade of tumor?—a
Thoeny HC. Diffusion-weighted MR imag-
tumor from treatment effect in patients meta-analysis. AJR Am J Roentgenol 2013;
ing including bi-exponential fitting for the
with glioblastoma: initial clinical experi- 200(3):624–629.
detection of recurrent or residual tumour
ence. AJNR Am J Neuroradiol 2014;35(3):
95. Waldman AD, Jackson A, Price SJ, et al. after (chemo)radiotherapy for laryngeal
490–497.
Quantitative imaging biomarkers in neuro- and hypopharyngeal cancers. Eur Radiol
84. Chandarana H, Kang SK, Wong S, et al. oncology. Nat Rev Clin Oncol 2009;6(8): 2013;23(2):562–569.
Diffusion-weighted intravoxel incoherent 445–454.
motion imaging of renal tumors with his- 106. Barral M, Taouli B, Guiu B, et al. Diffusion-
topathologic correlation. Invest Radiol 96. Thoeny HC, De Keyzer F, King AD. Diffu- weighted MR imaging of the pancreas: cur-
2012;47(12):688–696. sion-weighted MR imaging in the head and rent status and recommendations. Radiol-
neck. Radiology 2012;263(1):19–32. ogy 2015;274(1):45–63.
85. Fujima N, Yoshida D, Sakashita T, et al. In-
travoxel incoherent motion diffusion-weight- 97. Koyasu S, Iima M, Umeoka S, et al. The 107. Lee SS, Byun JH, Park BJ, et al. Quanti-
ed imaging in head and neck squamous cell clinical utility of reduced-distortion read- tative analysis of diffusion-weighted mag-
carcinoma: assessment of perfusion-relat- out-segmented echo-planar imaging in the netic resonance imaging of the pancreas:
ed parameters compared to dynamic con- head and neck region: initial experience. usefulness in characterizing solid pancre-
trast-enhanced MRI. Magn Reson Imaging Eur Radiol 2014;24(12):3088–3096. atic masses. J Magn Reson Imaging 2008;
2014;32(10):1206–1213. 28(4):928–936.
98. Sumi M, Van Cauteren M, Sumi T, Obara
86. Suo S, Lin N, Wang H, et al. Intravoxel M, Ichikawa Y, Nakamura T. Salivary gland 108. Re TJ, Lemke A, Klauss M, et al. Enhanc-
incoherent motion diffusion-weighted MR tumors: use of intravoxel incoherent mo- ing pancreatic adenocarcinoma delineation
imaging of breast cancer at 3.0 tesla: com- tion MR imaging for assessment of diffu- in diffusion derived intravoxel incoherent
parison of different curve-fitting methods. J sion and perfusion for the differentiation of motion f-maps through automatic vessel
Magn Reson Imaging 2015;42(2):362–370. benign from malignant tumors. Radiology and duct segmentation. Magn Reson Med
87. Padhani AR, Koh DM, Collins DJ. Whole- 2012;263(3):770–777. 2011;66(5):1327–1332.
body diffusion-weighted MR imaging in
99. Sumi M, Nakamura T. Head and neck tu- 109. Klauss M, Lemke A, Grünberg K, et al. In-
cancer: current status and research direc-
mours: combined MRI assessment based travoxel incoherent motion MRI for the dif-
tions. Radiology 2011;261(3):700–718.
on IVIM and TIC analyses for the differ- ferentiation between mass forming chronic
88. Yamasaki F, Kurisu K, Satoh K, et al. entiation of tumors of different histological pancreatitis and pancreatic carcinoma. In-
Apparent diffusion coefficient of human types. Eur Radiol 2014;24(1):223–231. vest Radiol 2011;46(1):57–63.

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 29

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

110. Concia M, Sprinkart AM, Penner AH, et ing diffusion-weighted intravoxel incoher- ciety for Magnetic Resonance in Medicine,
al. Diffusion-weighted magnetic resonance ent motion imaging. Magn Reson Imaging 2014.
imaging of the pancreas: diagnostic benefit 2011;29(8):1053–1058.
132. Nogueira L, Brandão S, Matos E, et al.
from an intravoxel incoherent motion mod-
122. Mazaheri Y, Vargas HA, Akin O, Goldman Application of the diffusion kurtosis model
el-based 3 b-value analysis. Invest Radiol
DA, Hricak H. Reducing the influence of b- for the study of breast lesions. Eur Radiol
2014;49(2):93–100.
value selection on diffusion-weighted imag- 2014;24(6):1197–1203.
111. Kang KM, Lee JM, Yoon JH, Kiefer B, Han ing of the prostate: evaluation of a revised
133. Rosen PP. Rosen’s breast pathology. Phila-
JK, Choi BI. Intravoxel incoherent motion monoexponential model within a clinical
delphia, Pa: Lippincott Williams & Wilkins,
diffusion-weighted MR imaging for charac- setting. J Magn Reson Imaging 2012;35(3):
2009; 194–195.
terization of focal pancreatic lesions. Radi- 660–668.
ology 2014;270(2):444–453. 134. Norddin N, Power C, Watson G, et al. Micro-
123. Riches SF, Hawtin K, Charles-Edwards EM,
112. Johnson LM, Turkbey B, Figg WD, Choyke scopic diffusion properties of fixed breast tis-
de Souza NM. Diffusion-weighted imaging
PL. Multiparametric MRI in prostate can- sue: preliminary findings. Magn Reson Med
of the prostate and rectal wall: compari-
cer management. Nat Rev Clin Oncol 2014 Dec 17. [Epub ahead of print]
son of biexponential and monoexponential
2014;11(6):346–353. modelled diffusion and associated perfu- 135. Cheng L, Bai Y, Zhang J, et al. Optimiza-
sion coefficients. NMR Biomed 2009;22(3): tion of apparent diffusion coefficient mea-
113. Rosenkrantz AB, Sigmund EE, Johnson G,
318–325. sured by diffusion-weighted MRI for diag-
et al. Prostate cancer: feasibility and pre-
liminary experience of a diffusional kurto- nosis of breast lesions presenting as mass
124. Shinmoto H, Tamura C, Soga S, et al. An
sis model for detection and assessment of and non-mass-like enhancement. Tumour
intravoxel incoherent motion diffusion-
aggressiveness of peripheral zone cancer. Biol 2013;34(3):1537–1545.
weighted imaging study of prostate cancer.
Radiology 2012;264(1):126–135. AJR Am J Roentgenol 2012;199(4):W496– 136. Iima M, Le Bihan D, Okumura R, et al. Ap-
114. Mazzoni LN, Lucarini S, Chiti S, Busoni W500. parent diffusion coefficient as an MR imag-
S, Gori C, Menchi I. Diffusion-weighted ing biomarker of low-risk ductal carcinoma
125. Dorrius MD, Dijkstra H, Oudkerk M, Si-
signal models in healthy and cancerous in situ: a pilot study. Radiology 2011;260(2):
jens PE. Effect of b value and pre-admission
peripheral prostate tissues: comparison of 364–372.
of contrast on diagnostic accuracy of 1.5-T
outcomes obtained at different b-values. J breast DWI: a systematic review and me- 137. Kwee TC, Takahara T, Luijten PR, Niev-
Magn Reson Imaging 2014;39(3):512–518. ta-analysis. Eur Radiol 2014;24(11):2835– elstein RA. ADC measurements of lymph
115. Shinmoto H, Oshio K, Tanimoto A, et al. 2847. nodes: inter- and intra-observer reproduc-
Biexponential apparent diffusion coeffi- ibility study and an overview of the litera-
126. Liu C, Liang C, Liu Z, Zhang S, Huang B.
cients in prostate cancer. Magn Reson Im- ture. Eur J Radiol 2010;75(2):215–220.
Intravoxel incoherent motion (IVIM) in
aging 2009;27(3):355–359. evaluation of breast lesions: comparison 138. Scaranelo AM, Eiada R, Jacks LM, Kulkarni
116. Shinmoto H, Oshio K, Tamura C, et al. with conventional DWI. Eur J Radiol 2013; SR, Crystal P. Accuracy of unenhanced MR
Diffusion-weighted imaging of prostate 82(12):e782–e789. imaging in the detection of axillary lymph
cancer using a statistical model based on node metastasis: study of reproducibility and
127. Bokacheva L, Kaplan JB, Giri DD, et al.
the gamma distribution. J Magn Reson Im- reliability. Radiology 2012;262(2):425–434.
Intravoxel incoherent motion diffusion-
aging 2015;42(1):56–62. weighted MRI at 3.0 T differentiates malig- 139. Fornasa F, Nesoti MV, Bovo C, Bonavina
117. Kuru TH, Roethke MC, Stieltjes B, et al. nant breast lesions from benign lesions and MG. Diffusion-weighted magnetic reso-
Intravoxel incoherent motion (IVIM) diffu- breast parenchyma. J Magn Reson Imaging nance imaging in the characterization of
sion imaging in prostate cancer: what does 2014;40(4):813–823. axillary lymph nodes in patients with breast
it add? J Comput Assist Tomogr 2014; cancer. J Magn Reson Imaging 2012;36(4):
128. Sigmund EE, Cho GY, Kim S, et al. Intravox-
38(4):558–564. 858–864.
el incoherent motion imaging of tumor mi-
118. Pang Y, Turkbey B, Bernardo M, et al. In- croenvironment in locally advanced breast 140. Kamitani T, Hatakenaka M, Yabuuchi H,
travoxel incoherent motion MR imaging for cancer. Magn Reson Med 2011;65(5):1437– et al. Detection of axillary node metastasis
prostate cancer: an evaluation of perfusion 1447. using diffusion-weighted MRI in breast can-
fraction and diffusion coefficient derived cer. Clin Imaging 2013;37(1):56–61.
129. Wu D, Li G, Zhang J, Chang S, Hu J, Dai
from different b-value combinations. Magn
Y. Characterization of breast tumors using 141. Luo N, Su D, Jin G, et al. Apparent dif-
Reson Med 2013;69(2):553–562.
diffusion kurtosis imaging (DKI). PLoS fusion coefficient ratio between axillary
119. Zhang YD, Wang Q, Wu CJ, et al. The his- One 2014;9(11):e113240. lymph node with primary tumor to detect
togram analysis of diffusion-weighted intra- nodal metastasis in breast cancer patients.
130. Tamura T, Usui S, Murakami S, et al. Bi-
voxel incoherent motion (IVIM) imaging for J Magn Reson Imaging 2013;38(4):824–
exponential signal attenuation analysis of
differentiating the gleason grade of pros- 828.
diffusion-weighted imaging of breast. Magn
tate cancer. Eur Radiol 2015;25(4):994–
Reson Med Sci 2010;9(4):195–207. 142. Gould MK, Fletcher J, Iannettoni MD, et
1004.
al. Evaluation of patients with pulmonary
131. Iima M, Kataoka M, Umehana M, et al.
120. Anderson AW, Gore JC. Analysis and correc- nodules: when is it lung cancer?: ACCP
Computer-assisted diagnosis of breast
tion of motion artifacts in diffusion weighted evidence-based clinical practice guidelines
lesions based on IVIM and non Gaussian
imaging. Magn Reson Med 1994;32(3): (2nd edition). Chest 2007;132(3 Suppl):
diffusion MRI [abstr]. In: Proceedings of
379–387. 108S–130S.
the Twenty-Second Meeting of the Interna-
121. Döpfert J, Lemke A, Weidner A, Schad tional Society for Magnetic Resonance in 143. Liu H, Liu Y, Yu T, Ye N, Wang Q. Eval-
LR. Investigation of prostate cancer us- Medicine. Berkeley, Calif: International So- uation of apparent diffusion coefficient

30 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

associated with pathological grade of evaluation with intravoxel incoherent mo- 164. Koh DM, Scurr E, Collins DJ, et al. Co-
lung carcinoma, before therapy. J Magn tion echo-planar MR imaging. Radiology lorectal hepatic metastases: quantitative
Reson Imaging 2014 Dec 24. [Epub ahead 1999;210(3):617–623. measurements using single-shot echo-pla-
of print] nar diffusion-weighted MR imaging. Eur
155. Luciani A, Vignaud A, Cavet M, et al.
Radiol 2006;16(9):1898–1905.
144. Reischauer C, Froehlich JM, Pless M, Liver cirrhosis: intravoxel incoherent mo-
Binkert CA, Koh DM, Gutzeit A. Early tion MR imaging—pilot study. Radiology 165. Joo I, Lee JM, Yoon JH, Jang JJ, Han JK,
treatment response in non-small cell lung 2008;249(3):891–899. Choi BI. Nonalcoholic fatty liver disease:
cancer patients using diffusion-weighted im- 156. Patel J, Sigmund EE, Rusinek H, Oei M, intravoxel incoherent motion diffusion-
aging and functional diffusion maps: a feasi- Babb JS, Taouli B. Diagnosis of cirrhosis weighted MR imaging-an experimental study
bility study. PLoS One 2014;9(10):e108052. with intravoxel incoherent motion diffu- in a rabbit model. Radiology 2014;270(1):
sion MRI and dynamic contrast-enhanced 131–140.
145. Chen L, Zhang J, Bao J, et al. Meta-anal-
ysis of diffusion-weighted MRI in the dif- MRI alone and in combination: prelimi- 166. Van Beers BE. Science to Practice: can we
ferential diagnosis of lung lesions. J Magn nary experience. J Magn Reson Imaging diagnose nonalcoholic steatohepatitis with
Reson Imaging 2013;37(6):1351–1358. 2010;31(3):589–600. intravoxel incoherent motion diffusion-
157. Dyvorne HA, Galea N, Nevers T, et al. weighted MR imaging? Radiology 2014;270
146. Li B, Li Q, Chen C, Guan Y, Liu S. A sys-
Diffusion-weighted imaging of the liver (1):1–2.
tematic review and meta-analysis of the
accuracy of diffusion-weighted MRI in the with multiple b values: effect of diffusion
167. Rapacchi S, Wen H, Viallon M, et al. Low
detection of malignant pulmonary nodules gradient polarity and breathing acquisition
b-value diffusion-weighted cardiac mag-
and masses. Acad Radiol 2014;21(1):21–29. on image quality and intravoxel incoherent
netic resonance imaging: initial results in
motion parameters—a pilot study. Radiol-
147. Halavaara J, Breuer J, Ayuso C, et al. humans using an optimal time-window im-
ogy 2013;266(3):920–929.
Liver tumor characterization: comparison aging approach. Invest Radiol 2011;46(12):
158. Yoon JH, Lee JM, Yu MH, Kiefer B, Han 751–758.
between liver-specific gadoxetic acid diso-
JK, Choi BI. Evaluation of hepatic focal
dium-enhanced MRI and biphasic CT—a 168. Delattre BM, Viallon M, Wei H, et al. In
lesions using diffusion-weighted MR imag-
multicenter trial. J Comput Assist Tomogr vivo cardiac diffusion-weighted magnetic
ing: comparison of apparent diffusion co-
2006;30(3):345–354. resonance imaging: quantification of nor-
efficient and intravoxel incoherent motion-
148. Bipat S, van Leeuwen MS, Comans EF, et derived parameters. J Magn Reson Imaging mal perfusion and diffusion coefficients
al. Colorectal liver metastases: CT, MR im- 2014;39(2):276–285. with intravoxel incoherent motion imaging.
aging, and PET for diagnosis—meta-analy- Invest Radiol 2012;47(11):662–670.
159. Kim YK, Lee MW, Lee WJ, et al. Diag-
sis. Radiology 2005;237(1):123–131. 169. Baur A, Stäbler A, Brüning R, et al. Dif-
nostic accuracy and sensitivity of diffusion-
149. Koh DM, Collins DJ. Diffusion-weighted weighted and of gadoxetic acid-enhanced fusion-weighted MR imaging of bone mar-
MRI in the body: applications and chal- 3-T MR imaging alone or in combination row: differentiation of benign versus path-
lenges in oncology. AJR Am J Roentgenol in the detection of small liver metastasis ologic compression fractures. Radiology
2007;188(6):1622–1635. ( 1.5 cm in diameter). Invest Radiol 1998;207(2):349–356.
2012;47(3):159–166. 170. Herneth AM, Philipp MO, Naude J, et al.
150. Lee Y, Lee SS, Kim N, et al. Intravoxel
incoherent motion diffusion-weighted MR 160. Chung WS, Kim MJ, Chung YE, et al. Vertebral metastases: assessment with ap-
imaging of the liver: effect of triggering Comparison of gadoxetic acid-enhanced parent diffusion coefficient. Radiology 2002;
methods on regional variability and mea- dynamic imaging and diffusion-weighted 225(3):889–894.
surement repeatability of quantitative pa- imaging for the preoperative evaluation of
171. Padhani AR, Makris A, Gall P, Collins DJ,
rameters. Radiology 2015;274(2):405–415. colorectal liver metastases. J Magn Reson
Tunariu N, de Bono JS. Therapy moni-
Imaging 2011;34(2):345–353.
151. Taouli B, Sandberg A, Stemmer A, et al. toring of skeletal metastases with whole-
Diffusion-weighted imaging of the liver: 161. Holzapfel K, Eiber MJ, Fingerle AA, Brue- body diffusion MRI. J Magn Reson Imaging
comparison of navigator triggered and gel M, Rummeny EJ, Gaa J. Detection, 2014;39(5):1049–1078.
breathhold acquisitions. J Magn Reson Im- classification, and characterization of focal
liver lesions: value of diffusion-weighted MR 172. Ichikawa S, Motosugi U, Ichikawa T, Sano
aging 2009;30(3):561–568.
imaging, gadoxetic acid-enhanced MR imag- K, Morisaka H, Araki T. Intravoxel inco-
152. Kandpal H, Sharma R, Madhusudhan KS, ing and the combination of both methods. herent motion imaging of the kidney: alter-
Kapoor KS. Respiratory-triggered versus Abdom Imaging 2012;37(1):74–82. ations in diffusion and perfusion in patients
breath-hold diffusion-weighted MRI of liver with renal dysfunction. Magn Reson Imag-
lesions: comparison of image quality and 162. Koh DM, Collins DJ, Wallace T, Chau I, ing 2013;31(3):414–417.
apparent diffusion coefficient values. AJR Riddell AM. Combining diffusion-weighted
MRI with Gd-EOB-DTPA-enhanced MRI im- 173. Chandarana H, Lee VS, Hecht E, Taouli B,
Am J Roentgenol 2009;192(4):915–922.
proves the detection of colorectal liver me- Sigmund EE. Comparison of biexponential
153. Chen X, Qin L, Pan D, et al. Liver diffusion- tastases. Br J Radiol 2012;85(1015):980– and monoexponential model of diffusion
weighted MR imaging: reproducibility com- 989. weighted imaging in evaluation of renal
parison of ADC measurements obtained lesions: preliminary experience. Invest Ra-
163. Taouli B, Vilgrain V, Dumont E, Daire JL,
with multiple breath-hold, free-breathing, re- diol 2011;46(5):285–291.
Fan B, Menu Y. Evaluation of liver diffu-
spiratory-triggered, and navigator-triggered
sion isotropy and characterization of focal 174. Gaing B, Sigmund EE, Huang WC, et al.
techniques. Radiology 2014;271(1):113–125.
hepatic lesions with two single-shot echo- Subtype differentiation of renal tumors
154. Yamada I, Aung W, Himeno Y, Nakagawa planar MR imaging sequences: prospective using voxel-based histogram analysis of
T, Shibuya H. Diffusion coefficients in study in 66 patients. Radiology 2003;226 intravoxel incoherent motion parameters.
abdominal organs and hepatic lesions: (1):71–78. Invest Radiol 2015;50(3):144–152.

Radiology: Volume 278: Number 1—January 2016 n radiology.rsna.org 31

REVIEW: Clinical Intravoxel Incoherent Motion and Diffusion MR Imaging Iima and Le Bihan

175. Zhang JL, Sigmund EE, Rusinek H, et al. 178. Cohen AD, Schieke MC, Hohenwalter MD, in breast cancer. Magn Reson Med 2014
Optimization of b-value sampling for diffu- Schmainda KM. The effect of low b-values Oct 9. [Epub ahead of print]
sion-weighted imaging of the kidney. Magn on the intravoxel incoherent motion de-
181. Iima M, Yamamoto A, Le Bihan D, et al.
Reson Med 2012;67(1):89–97. rived pseudodiffusion parameter in liver.
Computer-aided diagnosis of head and
Magn Reson Med 2014;73(1):306–311.
176. Lemke A, Stieltjes B, Schad LR, Laun FB. neck lesions from non-Gaussian diffusion
Toward an optimal distribution of b values 179. Jambor I, Merisaari H, Aronen HJ, et al. MRI signal patterns [abstr]. In: Proceed-
for intravoxel incoherent motion imaging. Optimization of b-value distribution for bi- ings of the Twenty-Third Meeting of the
Magn Reson Imaging 2011;29(6):766–776. exponential diffusion-weighted MR imaging International Society for Magnetic Reso-
of normal prostate. J Magn Reson Imaging nance in Medicine. Berkeley, Calif: Inter-
177. Leporq B, Saint-Jalmes H, Rabrait C, et
2014;39(5):1213–1222. national Society for Magnetic Resonance in
al. Optimization of intra-voxel incoherent
Medicine, 2015: 0687.
motion imaging at 3.0 Tesla for fast liver 180. Cho GY, Moy L, Zhang JL, et al. Compari-
examination. J Magn Reson Imaging 2015; son of fitting methods and b-value sampling
41(5):1209–1217. strategies for intravoxel incoherent motion

32 radiology.rsna.org n Radiology: Volume 278: Number 1—January 2016