See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/322246773

Short diffusion time DWI with oscillating gradient preparation as an early MRI
biomarker for radiation therapy response monitoring in glioblastoma: A pre-
clinical feasibility study

Article  in  International Journal of Radiation OncologyBiologyPhysics · January 2018

DOI: 10.1016/j.ijrobp.2017.12.280

CITATIONS READS

5 228

3 authors, including:

Eric K C Hau Han Shen

Crown Princess Mary Westmead Cancer Care Centre Westmead Institute for Medical Research
37 PUBLICATIONS   229 CITATIONS    35 PUBLICATIONS   136 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Radiosensitisation of Diffuse Intrinsic Pontine Glioma (DIPG) cells by modulating glucose metabolism View project

All content following this page was uploaded by Han Shen on 05 April 2018.

The user has requested enhancement of the downloaded file.

 International Journal of
Radiation Oncology
biology physics

www.redjournal.org

Physics Contribution

Short Diffusion Time Diffusion-Weighted

Imaging With Oscillating Gradient Preparation as
an Early Magnetic Resonance Imaging Biomarker
for Radiation Therapy Response Monitoring in
Glioblastoma: A Preclinical Feasibility Study
Andre Bongers, PhD,* Eric Hau, MD, PhD,y,z and Han Shen, PhDy
*Biological Resources Imaging Laboratory, Mark Wainwright Analytical Centre, The University of
New South Wales, Sydney, Australia; yRadiation Biology Group, Centre for Cancer Research, The
Westmead Institute for Medical Research, Sydney, Australia; and zThe Crown Princess Mary Cancer
Centre, Westmead Hospital, Sydney, Australia

Received Sep 8, 2017, and in revised form Nov 21, 2017. Accepted for publication Dec 19, 2017.

Summary Purpose: To investigate a novel alternative diffusion-weighted imaging (DWI)

This study investigates DWI approach using oscillating gradients preparation (OGSE) to obtain much shorter effec-
at very short diffusion times tive diffusion times (Deff) for tumor response monitoring by apparent diffusion coef-
(Deff) as a novel tool for ra- ficient (ADC) mapping in a glioblastoma mouse model.
diation therapy monitoring Methods and Materials: Twenty-four BALB/c nude mice inoculated with U87
and response detection in a glioblastoma cells were randomized into a control group and an irradiation group, which
glioblastoma mouse model. underwent a 15-day fractioned radiation therapy (RT) course with 2 Gy/d. Therapy
Using oscillating gradient response was assessed by mapping of ADCs at 6 time points using an in-house implemen-
diffusion MRI (OGSE) to tation of a cos-OGSE DWI sequence with Deff Z 1.25 ms and compared with a standard
overcome Deff limits in con- pulsed gradient DWI protocol (PGSE) with typical clinical diffusion time Deff Z 18 ms.
ventional DWI, we show that Longitudinal ADC changes in tumor and contralateral white matter (WM) were statisti-
ADC mapping at short Deff cally assessed using repeated-measures analysis of variance and post hoc (Sidak) testing.
can detect radiation changes Results: On short Deff OGSE maps tumor ADC was generally 30%-50% higher than in
significantly earlier and more surrounding WM. Areas correlated well with histology. Tumor identification was gener-
robustly than standard clin- ally more difficult on PGSE maps owing to nonsignificant WM/tumor contrast. During
ical DWI. We believe that RT, OGSE maps also showed significant tumor ADC increase (approximately 15%) in
short Deff DWI is highly response to radiation, consistently seen after 14-Gy RT dose. The clinical reference
promising for radiation (PGSE) showed lower sensitivity to radiation changes, and no significant response across
the radiation group and time course could be detected.

Reprint requests to: Andre Bongers, PhD, The University of New South Supplementary material for this article can be found at www
Wales, Biological Resources Imaging Laboratory, Mark Wainwright .redjournal.org.
Analytical Centre, UNSW Sydney, Sydney 2052, Australia. Tel: (þ61) AcknowledgmentsdThe authors thank the National Imaging Facility at
451015799; E-mail: [email protected] the UNSW Mark Wainwright Analytical Centre, Biological Imaging
Conflict of interest: none. Resources Laboratory, for the facilities and for scientific and technical
assistance.

Int J Radiation Oncol Biol Phys, Vol. -, No. -, pp. 1e10, 2018
0360-3016/$ - see front matter Ó 2018 Elsevier Inc. All rights reserved.
https://doi.org/10.1016/j.ijrobp.2017.12.280
2 Bongers et al. International Journal of Radiation Oncology  Biology  Physics

therapy monitoring on a Conclusion: Our short Deff DWI method using OGSE better reflected histologically
cellular level. defined tumor areas and enabled more consistent and earlier detection of microstructural
radiation changes than conventional methods. Oscillating gradients preparation offers sig-
nificant potential as a robust microstructural RT response biomarker, potentially helping to
shift important therapy decisions to earlier stages in the RT time course. Ó 2018 Elsevier
Inc. All rights reserved.

Introduction relatively long diffusion times with moderate diffusion

weightings. In this setting tissue-bound water will always
encounter restrictions on multiple length scales arising
High-grade gliomas are the most common primary ma-
from various internal and external structures (eg, nuclei,
lignant central nervous system tumors in adults, ac-
counting for nearly 80% of all cases (1). Radiation membranes, cytosol), which may lead to a loss of selec-
tivity in the relevant length range.
therapy (RT) has been the cornerstone of treatment for
Although extended experimental DWI protocols with
decades, and its benefit has been demonstrated in
multiple and/or strong diffusion weighting have been able
numerous clinical trials (2). For assessing high-grade
to better model restricted diffusion and reveal subtle dif-
gliomas response to therapy, the most widely used im-
ferences in the complicated tumor environment (24), these
aging criteria currently are based on 2-dimensional tumor
are generally too time-consuming to be useful in a clinical
measurements on anatomic computed tomography or
setting. Similarly, attempts to increase the predictive power
magnetic resonance imaging (MRI), in conjunction with
clinical assessment (3). However, macroscopic changes in of DWI for radiation treatment by using very strong
diffusion weighting (20, 23) have been somewhat suc-
tumor size can only be reliably detected in late stages of
cessful but severely suffer from low signal-to-noise ratio
treatment (usually 8-10 weeks after its start), and mea-
and long scan times.
surements are often not a true surrogate of tumor response
A promising alternative approach is guided by the idea
or patient survival (3). To effectively guide patient-
that DWI at very short diffusion times in the low milli-
tailored treatment strategies (4), it is desirable to
second range selectively samples subcellular diffusion
develop imaging tools to assess microstructural changes
length scales and may therefore be able to pick up micro-
much earlier in the treatment course.
A promising modality with this capability is diffusion- structural changes in internal cell structures more selec-
tively. Unfortunately, the smallest diffusion times achievable
weighted imaging (DWI) (5), which derives its image
with conventional clinical DWI protocols are strongly
contrast from water molecules that randomly probe intra-
restricted by available gradient hardware, so that subcellular
and extracellular barriers. Mapping of apparent diffusion
diffusion length is hard to achieve on (pre-)clinical scanners.
coefficients (ADCs) with dedicated pulse sequences yields
However, recently novel DWI protocols using oscillating
valuable information about such restrictions and has been
gradient preparation (OGSE) have been proposed (25, 26),
used to detect and model microstructural features in various
which are able to circumvent these limitations using diffu-
organs and diseases (6-9). Specifically, it has been found
that ADC strongly correlates with cell density in brain tu- sion gradients in periodic wave-trains instead of standard
pulsed gradient preparation (27).
mors (10-12).
Initial computer simulations (28) and preclinical chemo-
On the basis of the hypothesis that therapy-induced cell
therapy studies (29) indicate that such methods may be able
damage leads to changes in water mobility well before
to substantially increase sensitivity of DWI to subtle
macroscopic changes in tumor morphology can be detected,
microstructural changes in tumors. Although from this it
several studies have shown that ADC mapping has merit as
may be anticipated that such methods could be specifically
an early predictor of therapeutic efficacy in cancer treat-
promising as an imaging early biomarker for radiation
ment (13-17).
Diffusion-weighted imaging has also shown some response detection, no studies to date have investigated their
potential to monitor cellular damage in a fractioned RT
promising initial results for RT response monitoring (18-
setting. In this study we investigated the value of DWI at
22), but unfortunately it was often discovered that the
short (low millisecond range) diffusion times for the detec-
detection sensitivity of conventional DWI methods is too
tion of microstructural tissue changes and therapy response
low to be useful for widespread clinical therapy monitoring,
detection in a preclinical glioblastoma mouse model. We use
for example to clearly indicate radiation response in indi-
an in-house implementation of an OGSE sequence variant
vidual subjects (20, 21, 23) or to reliably differentiate be-
(26) to longitudinally monitor a fractioned RT protocol
tween treatment options (22).
A particular reason for this may be a loss of sensitivity through ADC mapping and statistically compare the results
with a standard clinical protocol using pulsed gradient
through an averaging effect: because of technical limita-
diffusion preparation (PGSE) and parameters.
tions conventional clinical DWI methods usually use
Volume -  Number -  2018 Short diffusion time DWI for radiation therapy monitoring 3

Methods and Materials growth and an earlier endpoint after day 25. For histologic
evaluation 1 additional mouse per time point per group was
Animal model and radiation treatment scanned, followed by brain harvest for hematoxylin and
eosin (H&E) staining and histologic evaluation.
All experiments were performed in an orthotopic U87
glioblastoma mouse model, with all animal procedures DWI MRI protocol
approved by the local Animal Care and Ethics Committee
(ACEC) of the University of New South Wales, Australia. All imaging was performed on a 9.4T BioSpec Avance III
Briefly, 24 female athymic nude mice (BALB/c, 8 weeks) 94/20 magnetic resonance micro-imaging system (Bruker
were intracranially stereotactically injected with 5  104 BioSpin GmbH, Germany) equipped with a 23-mm mouse
U87 cells in the right caudate putamen using the following head volume radiofrequency coil and BGA-12S
coordinates: 1 mm anterior, 1.5 mm lateral, and 3.0 mm HP gradients with gradient strength 660 mT/m and slew
below the bregma. rate 4570 Tm/s (Bruker BioSpin GmbH, Germany).
Tumor inoculated mice were randomly divided into a For each time point mice underwent an identical MRI
control group and a radiation group. The control group protocol, using 3 consecutive pulse sequences: (1) an in-house
consisted of 10 mice, which were retained for MRI only, developed OGSE method to specifically sample short diffu-
without any further treatment. The radiation group (12 mice) sion times, (2) a conventional PGSE protocol (27) as typically
underwent fractioned RT administered in 24-hour intervals used in clinical applications, and (3) a standard T2-weighted
starting at day 13 after tumor inoculation, with 2 Gy per TurboRARE scan for anatomic reference. The overall imag-
fraction on 15 consecutive days. Total administered dose was ing protocol took approximately 50 minutes to complete.
30 Gy. The dose administration schedule is summarized The general features and specific diffusion parameters
together with the MRI scanning schedule in Table 1. for the DWI pulse sequences (1) and (2) in the imaging
For each radiation fraction mice were immobilized inside protocol are compared in Table 2. Key to reducing
a custom built lead box to selectively expose the entire brain achievable diffusion time is the replacement of the con-
while shielding the rest of the body. The shield box was ventional Steijskal-Tanner diffusion preparation (27) with a
designed to hold a group of up to 5 mice at a time and cosine-shaped wave train in the cos-OGSE sequence. As a
equipped with a nose cone for gas anesthesia. The irradiation result the diffusion time only depends on the achievable
setup is illustrated in the Supplementary Material (Fig. S1; oscillation frequency of the gradient wave train (25). This
available online at www.redjournal.org). Before radiation overcomes limitationsdposed by the gradient hardwared
treatment the animals were anesthetized with a 2% iso- that arise from the strong interdependence between b-value
flourane/oxygen mixture at a flow rate of 1 L/min. Radiation and diffusion time in PGSE (see b-value/diffusion time
with an energy of 320 kV was then administered using a self- relationships in Table 2). On our preclinical system OGSE
contained precision X-ray irradiation system, X-RAD 320 oscillation frequencies of 250 Hz were achievable at b-
(Precision X-Ray, North Branford, CT) with a 25-cm-wide values of up to 800 s/mm2. The specific DWI parameters
uniform radiation field that was adjusted to the mouse heads. used in our experiments can be found in Table 2.
Dose was monitored and verified using thermoluminscent For our experiments we used an OGSE frequency
dosimeters on the surface of a mouse simulating phantom f Z 200 Hz, close to the achievable maximum of our
inside the shielding and restraint apparatus. gradient system, which corresponds to an effective diffu-
Five mice from each group underwent consecutive MRI sion time of Deff Z 1.25 ms. The rationale for this choice
scans at multiple time points (Table 1) with a protocol as was from 2 theoretical considerations: first, previous
described below. Animals in the irradiation group received simulation studies (28) have shown that DWI sensitivity to
6 longitudinal MRI scans, whereas control group animals changes of cell microstructure is expected to increase with
underwent MRI at 5 time points owing to excessive tumor smaller diffusion times. We have therefore aimed to shorten

Table 1 MRI time schedule for control and treatment groups

Day after tumor injection
Group 11 15 20 22 25 36
Control group
Radiation dose (Gy) 0 0 0 0 0 Controls euthanized
No. of animals for MRI 5þ1 5þ1 5þ1 5þ1 5þ1 after day 25
Radiation group
Radiation dose (Gy) 0 4 14 18 24 30
No. of animals for MRI 5þ1 5þ1 5þ1 5þ1 5þ1 5þ1
Abbreviation: MRI Z magnetic resonance imaging.
5 animals in each group underwent the full MRI time-course comprising 5 or 6 timepoints, respectively. One additional animal (indicated as þ1) was
euthanized at each time-point and used for histological examination and replaced by a new animal for the next MRI.
4 Bongers et al. International Journal of Radiation Oncology  Biology  Physics

Table 2 Main features and experimental parameters of the short diffusion time OGSE and clinical PGSE methods
OGSE PGSE
Sequence diagram of
the DWI preparation
scheme with associated
parameter nomenclature

 
Effective diffusion time Deff Z 1
4fosc
Z 14$sosc Deff Z D  3d

B value bZ ð1  1=8NÞ
8p2
$ð g$G$sosc Þ2 $T b Z ðg$G$dÞ2 $Deff

Specific DWI Deff Z 1.25 ms, lfree Z 4 mm Deff Z 18 ms, lfree Z 15 mm

parameters as fosc Z 200 Hz, sosc Z 5 ms, d Z 5 ms,
used in experiment D Z 34 ms, T Z 30 ms D Z 20 ms,
bx Z by Z bz Z 600 s/mm2 bx Z by Z bz Z 600 s/mm2
Abbreviations: DWI Z diffusion-weighted imaging; OGSE Z oscillating gradient preparation; PGSE Z pulsed gradient diffusion preparation.

diffusion times as much as feasible while keeping the duty co-registered to the first ADC map in each time series using an
cycle of our gradient system in a reasonable range. Second, affine transformation with a Mattes mutual information metric.
the free diffusion length (from Einstein’s relation
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi For further processing the central tumor slice was extracted
hr 2 iZ 6$Deff $D0 , with typical D0 Z 3.0 mm2/ms for from each 3-dimensional ADC map volume and interpolated to
tissue water at 37 [30]) at this Deff should approximately match the T2-weighted anatomy images.
correspond to the typical cell radius (r w 5 mm) of our U87 Before statistical analysis diffusion maps with exces-
tumor model. Our (qualitative) hypothesis was that probing sively high ADC values across the entire brain (which may
water mobility in the range of expected changes should be caused by rigid-body motion of the mouse head) were
maximize the detection sensitivity. detected using a statistical method as described in the
The short diffusion time method was compared with a Supplementary Material (available online at www.
PGSE method with an effective diffusion time of redjournal.org). These rare occurrences (4 single time
Deff Z 18 ms, which corresponds to a value typically used in a points) were removed from the time courses. Regions of
clinical setting. To allow for direct comparison of the specific interest (ROIs) for tumor and symmetric contralateral
effect of diffusion time on radiation response sensitivity, both healthy tissue were then delineated on the first anatomic
methods implemented an identical spin-echo readout with T2-weighted image of each time series and transferred to
exactly matched sequence parameters: echo time (TE) all co-registered ADC maps to calculate mean ADC and
70.2ms, repetition time (TR) 2500ms, field of view standard deviation for each individual time point and
1.5  1.5 cm, matrix 64  64, resolution 234  234 mm, slice method. The delineation of tumor regions on the initial T2-
thickness: 2 mm þ 0.2 mm gap, 3 slices, 2 averages. weighted anatomy was chosen deliberately to guarantee
their independence of the DWI methods under investigation
and prevent double dipping and bias from multiple ROI
Image and data analysis delineations (34). Hereby, consistency of ROI alignment
with the tumor regions on the longitudinal ADC maps was
All data were postprocessed using in-house-developed Python ensured by the preceding co-registration procedure.
software, based on ITK (31, 32) and SciPy (33) data analysis
libraries. Averaged diffusion-weighted images were calculated
as the geometric mean from all orthogonal gradient directions. Statistical analysis
The ADC maps were calculated by fitting a mono-exponential
decay model S(b) Z S0$exp(b$ADC) to all pixels within a Radiation response in the mean ADC time course was
presegmented brain mask. To ensure reproducible alignment of statistically tested for each animal group and method using
the tumor areas on all longitudinal scans, all ADC maps were a 2-way repeated-measures analysis of variance (ANOVA).
Volume -  Number -  2018 Short diffusion time DWI for radiation therapy monitoring 5

Time series were tested for significant differences in the 1c and 1d, respectively, with corresponding ADC histo-
temporal ADC behavior between tumor and contralateral grams of tumor and contralateral WM ROIs in Figs. 1e and
healthy white matter (WM), by analyzing the interaction 1f. Visual comparison reveals that short diffusion time DWI
between the factors “tissue type” (ROI side) and “time delivers ADC maps with markedly higher tumor/WM
point” (radiation dose). An interaction P value PInteraction < contrast than the standard clinical reference protocol. This
103 was considered to prove H0 (no interaction) wrong, was generally the case for all investigated animals and re-
thus revealing a significant difference in temporal behavior gions: tumor areas could be easily identified in OGSE maps
in tumor and contralateral side. A Sidak multiple compar- and correlated well with H&E stained histology. Histo-
ison post hoc test was subsequently used to locate signifi- grams exhibited ADC differences in the range of 30%-50%,
cant ADC changes within the longitudinal series. For this, depending on the state of therapy [ADC(WM) w 0.7 mm2/s,
ADC at each time point of the respective series was ADC(Tumor) w 1.0-1.3 mm2/s]. On the other hand, on
compared with the baseline value (time point 1). A P value PGSE ADC maps, tumor/WM contrast was generally
PDADC < .05 was considered to reveal a significant increase poor, and histograms tended to largely overlap
in ADC. [ADC(WM) w 0.7 mm2/s, ADC(Tumor) w 0.6-0.7 mm2/s].
This was particularly true for control animals and for ani-
mals in early stages of treatment, making tumor delineation
Results on PGSE ADC maps very difficult in these cases.

Tumor delineation on ADC maps Longitudinal evaluation of ADC map series

Figure 1 shows ADC maps and histology of tumor-bearing Figure 2 shows the longitudinal ADC map evolution during
control (left) and irradiated (right) animals in late stages of tumor progression for an exemplary control (top) and irra-
tumor growth and RT. The ADC maps from short Deff diated animal (bottom). The OGSE and PGSE maps are
OGSE and reference PGSE protocols are compared in Figs. compared in Figs. 2b and 2c, respectively. It can be seen that

a b a b

c d c d

e Contralateral WM f Tumor
e f
30 30 30 Contralateral WM Tumor
Tumor 30
Contralateral WM Tumor Contralateral WM
relative count [%]
relative count [%]

relative count [%]

relative count [%]

25 25 25 25
20 20 20 20
15 15 15 15
10 10 10 10
5 5 5 5
0 0 0 0
0.6 0.8 1.0 1.2 1.4 1.6 0.6 0.8 1.0 1.2 1.4 1.6 0.6 0.8 1.0 1.2 1.4 1.6 0.6 0.8 1.0 1.2 1.4 1.6
ADC [μm/ms2] ADC [μm/ms2] ADC [μm/ms2] ADC [μm/ms2]

Fig. 1. Comparison of oscillating gradients preparation (OGSE) and pulsed gradient diffusion preparation (PGSE) apparent
diffusion coefficient (ADC) maps with anatomic magnetic resonance imaging and hematoxylin and eosinestained histology.
Left: Control animal, 20 days after tumor inoculation. Right: Irradiated animal after completed treatment (30-Gy dose). (a)
T2-Weighted anatomy; (b) hematoxylin and eosinestained histology slice; (c) OGSE ADC map [mm2/ms); (d) PGSE ADC
map; (e, f) ADC histogram from tumor and white matter region of interest for OGSE (e) and PGSE (f).
6 Bongers et al. International Journal of Radiation Oncology  Biology  Physics

Fig. 2. Longitudinal oscillating gradients preparation (OGSE) and pulsed gradient diffusion preparation (PGSE) apparent
diffusion coefficient (ADC) maps from typical control and radiation group animals. (a) T2-Weighted anatomy; (b) OGSE; (c)
PGSE.

OGSE ADC maps correlate well with tumor edema regions observations were consistent across all animals of our
on the T2-weighted anatomical images across all time points. experimental cohort: our OGSE experiment for the irradi-
Visual inspection of the control animal OGSE maps (Fig. 2b, ation group (Fig. 3a) exhibited an ADC increase in all
top) indicates that average tumor ADC values for this group animals, whichdin 4 of 5 animalsdconsistently occurred
largely remain constant during the entire time series. How- as a distinct step increase after a radiation dose of 14 Gy. A
ever, treated animals (Fig. 2b, bottom) show a clearly visible single animal showed a later response at approximately
increase in tumor ADC during the radiation time course. 24 Gy. No ADC changes were observed in contralateral
Conversely, conventional (PGSE) maps show later and less WM during fractioned RT. In control group animals
consistent ADC changes in response to radiation (Fig. 2c, (Fig. 3c) mean OGSE ADCs in both regions largely
bottom), making reliable detection of treatment effects remained constant across the entire time course, indicating
difficult. that, in our glioblastoma model, untreated tumor progres-
The time series plot of the ROI ADC averages in tumor sion has little or no influence on ADC. The time course of
and contralateral WM in Figure 3 shows that the above our reference PGSE protocol (Figs. 3b and 3d) shows much
Volume -  Number -  2018 Short diffusion time DWI for radiation therapy monitoring 7

a OGSE - Radtiation group b PGSE - Radtiation group

1.4 1.4
1.3 1.3
1.2 1.2
ADC [mm2/s]

ADC [mm2/s]
1.1 1.1
1.0 1.0
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 0 2 4 6 8 1012 14 16 18 20 22 24 26 28 30
Radiation Dose [Gy] Radiation Dose [Gy]

c OGSE - Control group d PGSE - Control group

1.4 1.4 Tumour, individual animal
1.3 1.3 Contralateral WM, individual animal
1.2 1.2 Tumour, cohort mean
ADC [mm2/s]

ADC [mm2/s]
1.1 1.1 Contralateral WM,
1.0 1.0 cohort mean
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
10 12 14 16 18 20 22 24 26 10 12 14 16 18 20 22 24 26
Time after tumor implant [days] Time after tumor implant [days]

Fig. 3. Longitudinal oscillating gradients preparation (OGSE) and pulsed gradient diffusion preparation (PGSE) apparent
diffusion coefficients (ADCs) for all examined animals. (a) OGSE irradiated group; (b) PGSE irradiated group; (c) OGSE
control group; (d) PGSE control group. Mean tumor ADC is shown in red, and white matter (WM) ADC in blue, with group
average in dark and individual animal time courses in light coloring. Error bars for individual animals indicate standard
deviation within each region of interest. Error bars in the group average indicate standard deviation between the region of
interest averages.

higher data variability between subjects and time points in but it may be classified as a general “all brain effect”
both (WM and tumor) ROIs, and no clear increase in tumor (possibly due to overall motion).
ADC was detectable with this protocol (Fig. 3d)
The statistical analysis of the longitudinal time courses
is detailed in Figure 4. The graphs show the grouped ADC Discussion
changes relative to the (untreated) baseline (t1) (DADC),
with ANOVA PInteraction values on top and post hoc P This work demonstrates that short diffusion time DWI (as
values at the bottom of each graph. Analysis of variance facilitated by novel OGSE methods) has significant po-
revealed that OGSE (as the only method) showed highly tential to improve monitoring of RT response. To the best of
significant variation between tumor and WM ADC time our knowledge this is the first study to investigate the value
courses in response to RT (Fig. 4a), indicating a significant of OGSE DWI to assess tumor response to RT. In our
DADC response to radiation for this method. Post hoc preclinical data OGSE seems to have some distinct ad-
(Sidak) testing indicated a consistently significant DADC vantages over clinically established conventional DWI
after time point 4 of our radiation treatment (14-Gy dose). methods. First, in our tumor model OGSE ADC maps
No statistically significant temporal radiation response exhibit stronger contrast between healthy tissue and tumor
could be found with the clinical reference PGSE protocol areas, enabling more reliable ADC-based delineation of
(Fig. 4b), owing to weaker DADC and high intersubject cancerous tissue. Second, OGSE detects significant
variation. Neither of the control groups exhibited signifi- microstructural changes after fractioned RT much earlier
cant ANOVA results (Figs. 4c and 4d). A slightly signif- and more reliably than conventional DWI. A first (pre-
icant post hoc result in the PGSE control group (Fig. 4d) liminary) analysis of high-resolution histologic data from
did not have a resemblance in significant ANOVA PInter- individual animals (see Supplementary Material, available
action and was therefore not considered as a tumor effect, online at www.redjournal.org) also shows that ADC
8 Bongers et al. International Journal of Radiation Oncology  Biology  Physics

a OGSE - Radiation group b PGSE - Radiation group

PInteraction = 4*10 (***)
-4 PInteraction= .3537

0.4 0.6

ΔADC {tn-t1} [mm2/s] 0.4

ΔADC {tn-t1} [mm2/s]

0.2
0.2

0.0 0.0

-0.2
-0.2
Tumour -0.4 Tumour
Contralateral WM Contralateral WM
-0.4 -0.6
t2-t1 t3-t1 t4-t1 t5-t1 t6-t1 t2-t1 t3-t1 t4-t1 t5-t1 t6-t1
Tumour 0.413 0.6904 0.0006 0.0012 < 0.0001 Tumour 0.727 0.878 0.103 0.099 0.516
PΔADC

PΔADC
WM 0.5683 > 0.999 0.9974 0.7716 > 0.999 WM 0.883 > 0.999 0.974 > 0.999 0.800

c d
OGSE - Control group PGSE - Control group
PInteraction= .9886 PInteraction= .7920
0.4 0.4
ΔADC {tn-t1} [mm2/s]

ΔADC {tn-t1} [mm2/s]

0.2 0.2

0.0 0.0

-0.2 -0.2
Tumour
Contralateral WM
-0.4 -0.4
t2-t1 t3-t1 t4-t1 t5-t1 t2-t1 t3-t1 t4-t1 t5-t1
Tumour 0.2554 0.9983 0.3954 0.4347 Tumour 0.5733 > 0.9999 0.1162 0.0236
PΔADC

PΔADC

WM 0.6106 > 0.9999 0.4554 0.3756 WM 0.7991 > 0.9999 0.5928 0.4672

Fig. 4. Summary of statistical analysis of apparent diffusion coefficient (ADC) changes in oscillating gradients preparation
(OGSE) and pulsed gradient diffusion preparation (PGSE) methods from untreated baseline (t1) in tumor (black) and white
matter (WM; gray). Box/whisker plots (25th/75th percentile, median, minimum/maximum) indicate the variation within the
respective group; the vertical lines next to each longitudinal point show the mean and 99% confidence interval. The result of
the repeated-measures two-way analysis of variance PInteraction for the mutual interaction between longitudinal changes in
contralateral and tumor side is reported above each plot. The P values of multi-comparison (Sidak) testing for differences
between baseline and later time points are tabulated below each graph, with significant responses indicated in red. (A color
version of this figure is available at www.redjournal.org.)

differences detected in OGSE correlate significantly better far longer than the average cell radius of rcell w 5 mm we
with obvious histologic features than conventional DWI. found in our histologic analysis. On the other hand, the
Specifically we found that therapy-induced changes in much shorter diffusion time Deff Z 1.25 ms in our OGSE
cellular components (cellular and nuclear sizes) visible on experiments
p ffiffiffiffiffiffiffiffi exhibit a free diffusion length of
H&E histology generally correlated well with ADC hr 2 i Z 4.5 mm and therefore samples subcellular
changes in short diffusion time scans. This correlation was changes more selectively. This may render the method
generally much weaker in the maps obtained with our more sensitive to the subtle changes in the specific length
conventional reference protocol. An explanation may be the scale of the tumor cellsdan interpretation that is, also
fact that, in the conventional DWI protocol, water mole- underpinned by numerical simulations, suggesting that
cules sample restrictions on multiple length scales during sensitivity to intracellular changes increases when diffusion
the time of signal evolution, which essentially leads to an is sampled on a length scale that closely matches cellular
averaging effect: the long diffusion time in the clinical dimensions (28). A previous preclinical chemotherapy
reference (PGSE) protocol (DeffpZ 18 ms) corresponds to
ffiffiffiffiffiffiffiffi study in a glioblastoma rat model also supports this notion
a free water diffusion radius of hr 2 i w 18 mm, which is (29), whereby increased ADC response could be detected in
Volume -  Number -  2018 Short diffusion time DWI for radiation therapy monitoring 9

the short diffusion time regime after treatment with the be expected for clinical OGSE. In our preclinical study we
commonly used chemothera-peutic drug 1,3-bis(2-chlor- have deliberately chosen to use a diffusion time close to the
oethyl)-1-nitrosourea (BCNU). lowest achievable value on our system (see Methods and
From these encouraging findings we strongly anticipate Materials). This decision was driven by the expectation from
that short diffusion time DWI also has a merit to signifi- previous simulation results and theoretical considerations
cantly improve sensitivity of DWI-based RT monitoring (28) that shorter diffusion times should increase sensitivity to
and response detection. subtle intracellular changes. If this hypothesis proves true in
In the preclincal domain, novel OGSE applications that practice, it would have to be expected that sensitivity of
have recently been explored to model microstructure of OGSE methods to radiation changes may be slightly reduced
cancerous tissues from DWI data may further add to this in a clinical setting compared with our results. However,
potential: using multi-frequency OGSE acquisitions (35, even on clinical scanners OGSE will still extend the clini-
36), sometimes in combination with conventional cally achievable diffusion time range to much lower values.
methods (37), such methods are able to map important In practice the optimum diffusion time will also likely
microstructural parameters such as cell sizes, cellularity, or depend on the tumor under investigation. With the specific
extracellular space in a quantitative way. The availability of aim of clinical translation in mind, in this this study we have
robustdand potentially quantitativedtreatment informa- only used a single diffusion time to keep the acquisition time
tion may help to better identify successful therapies with within a clinically feasible range. Future studies are planned
RT, chemotherapy, and combined treatment approaches in to explore a diffusion time range to correlate ADC response
preclinical experiments. with theoretical expectations and make use of quantitative
The translation of short diffusion time DWI to the clinical methods discussed above. Such studies will also help to
domain also has the significant potential to add more robust resolve a further limitation of the present study, which arises
diagnostic information about microstructural tissue changes from the fact that we have only used mean ADC values that
for cancer staging and treatment. This may help clinical de- were averaged over an entire tumor ROI as an “aggregate”
cision makers to shift important therapy decisions to much imaging biomarker to assess tumor response. However, in
earlier stages in the radiation course, for example enabling the agreement with many findings from studies with more
incorporation of alternative therapies when a lack of treatment conventional DWI acquisitions (21), our data generally
efficacy may be detected. Additional information about ADC showed significant ADC heterogeneity across tumor areas
heterogeneity across tumors that may be gained from such (see, eg, Figs. 1 and 2 and Supplementary Material, available
methods may also prove helpful as input for novel intensity online at www.redjournal.org). This variation was not
modulated RT approaches, such as dose painting by numbers investigated in detail in this study because our primary goal
(4, 38) (eg, enabling the application of a radiation boost to here was to explore the advantages of OGSE over standard
regions where suboptimal response is detected). When com- clinical DWI methods for longitudinal RT response detec-
bined with novel analysis techniques like texture analysis (39), tion. A detailed exploration of local ADC variations will
functional diffusion mapping (40), or radiomics approaches require more precise matching of the MRI acquisition with
(41), such heterogeneity analysis also has the potential to add histology data in a larger cohort. This investigation is under
valuable diagnostic information for tumor staging and clas- way and will, in conjunction with multi-frequency OGSE
sification. Although still in the research state, first translational experiments, allow much more detailed insights into the
studies using these methods have already shown benefits in biophysical basis of the ADC behavior in this methodology.
clinical oncology (eg, for cancer detection and classification in In the future this may potentially lead to the further estab-
the prostate [42, 43]). Applications in the detection of cancer lishment of highly sensitive, quantitative DWI biomarkers
response to treatment still remain to be explored. for response detection in RT.
Despite its anticipated potential, certain limitations have
to be considered when discussing the translation of short References
diffusion time DWI to clinical application. First, the feasi-
bility to assess the very short diffusion time regime as pro- 1. Louis DN, et al. The 2007 WHO classification of tumours of the
posed in our study is typically restricted by clinical gradient central nervous system. Acta Neuropathol 2007;114:97-109.
hardware when conventional DWI methods are used. To 2. Laperriere N, et al. Radiotherapy for newly diagnosed malignant
overcome these limitations, specific diffusion preparation glioma in adults: A systematic review. Radiother Oncol 2002;64:259-
schemes (such as OGSE in our study) will have to be used, 273.
3. Wen PY, et al. Updated response assessment criteria for high-grade
and dedicated methods will have to be developed on gliomas: Response assessment in neuro-oncology working group. J
research-enabled MRI scanners. Although it has been shown Clin Oncol 2010;28:1963-1972.
that this is feasible (44, 45), owing to weaker gradient 4. Bentzen SM. Dose painting and theragnostic imaging: Towards the
hardware, the achievable maximum oscillation frequencies prescription, planning and delivery of biologically targeted dose dis-
will be lower than in our preclinical experiments (ie, tributions in external beam radiation oncology. Cancer Treat Res
2008;139:41-62.
fosc < 100 Hz as opposed to fosc Z 200 Hz in our setting). 5. Le Bihan D. Apparent diffusion coefficient and beyond: What diffu-
Deffffi > 2.5 ms) and larger
As a result, longer diffusion timesp(ffiffiffiffiffiffiffi sion MR imaging can tell us about tissue structure. Radiology 2013;
free diffusion sampling range ð hr2 iw6 mmÞwill have to 268:318-322.
10 Bongers et al. International Journal of Radiation Oncology  Biology  Physics

6. Togao O, et al. Assessment of renal fibrosis with diffusion-weighted 26. Parsons EC, Does MD, Gore JC. Temporal diffusion spectroscopy:
MR imaging: Study with murine model of unilateral ureteral Theory and implementation in restricted systems using oscillating
obstruction 1. Radiology 2010;255:772-780. gradients. Magn Reson Med 2006;55:75-84.
7. Bourne R, et al. Effect of formalin fixation on biexponential modeling of 27. Stejskal E, Tanner J. Spin diffusion measurements: Spin echoes in the
diffusion decay in prostate tissue. Magn Reson Med 2013;70:1160-1166. presence of a time-dependent field gradient. J Chem Phys 1965;42:
8. Schlaug G, et al. Time course of the apparent diffusion coefficient 288.
(ADC) abnormality in human stroke. Neurology 1997;49:113-119. 28. Xu J, Does MD, Gore JC. Sensitivity of MR diffusion measurements
9. Guo Y, et al. Differentiation of clinically benign and malignant breast to variations in intracellular structure: Effects of nuclear size. Magn
lesions using diffusion-weighted imaging. J Magn Reson Imaging Reson Med 2009;61:828-833.
2002;16:172-178. 29. Colvin DC, et al. Earlier detection of tumor treatment response using
10. Ellingson BM, et al. Validation of functional diffusion maps (fDMs) as magnetic resonance diffusion imaging with oscillating gradients.
a biomarker for human glioma cellularity. J Magn Reson Imaging Magn Reson Imaging 2011;29:315-323.
2010;31:538-548. 30. Le Bihan D, Iima M. Diffusion magnetic resonance imaging: What
11. Gauvain KM, et al. Evaluating pediatric brain tumor cellularity with water tells us about biological tissues. PLoS Biol 2015;13:e1002203.
diffusion-tensor imaging. AJR Am J Roentgenol 2001;177:449-454. 31. Johnson HJ, McCormick MM, Ibanez L. The ITK Software Guide
12. Sugahara T, et al. Usefulness of diffusion-weighted MRI with echo- Book 1: Introduction and Development Guidelines. Clifton Park, New
planar technique in the evaluation of cellularity in gliomas. J Magn York: Kitware Inc; 2015.
Reson Imaging 1999;9:53-60. 32. Johnson HJ, McCormick MM, Ibanez L. The ITK Software Guide
13. Patterson DM, Padhani AR, Collins DJ. Technology insight: Water Book 2: Design and Functionality. Clifton Park, New York: Kitware
diffusion MRIda potential new biomarker of response to cancer Inc; 2015.
therapy. Nat Clin Pract Oncol 2008;5:220-233. 33. Oliphant TE. Python for scientific computing. Comput Sci Eng 2007;9:
14. Chenevert TL, McKeever PE, Ross BD. Monitoring early response of 10-20.
experimental brain tumors to therapy using diffusion magnetic reso- 34. Kriegeskorte N, et al. Circular analysis in systems neuroscience: The
nance imaging. Clin Cancer Res 1997;3:1457-1466. dangers of double dipping. Nat Neurosci 2009;12:535-540.
15. Hakumaki JM, et al. Quantitative 1H nuclear magnetic resonance 35. Jiang X, et al. In vivo imaging of cancer cell size and cellularity using
diffusion spectroscopy of BT4C rat glioma during thymidine kinase- temporal diffusion spectroscopy. Magn Reson Med 2017;78:156-164.
mediated gene therapy in vivo: Identification of apoptotic response. 36. Shemesh N, Álvarez GA, Frydman L. Measuring small compartment
Cancer Res 1998;58:3791-3799. dimensions by probing diffusion dynamics via Non-uniform Oscil-
16. Chenevert TL, et al. Diffusion magnetic resonance imaging: An early lating-Gradient Spin-Echo (NOGSE) NMR. J Magn Reson 2013;237:
surrogate marker of therapeutic efficacy in brain tumors. J Natl 49-62.
Cancer Inst 2000;92:2029-2036. 37. Reynaud O, et al. Pulsed and oscillating gradient MRI for assessment
17. Stegman LD, et al. Diffusion MRI detects early events in the response of cell size and extracellular space (POMACE) in mouse gliomas.
of a glioma model to the yeast cytosine deaminase gene therapy NMR Biomed 2016;29:1350-1363.
strategy. Gene Ther 2000;7:1005-1010. 38. Bentzen SM. Theragnostic imaging for radiation oncology: Dose-
18. Ross BD, et al. Evaluation of cancer therapy using diffusion magnetic painting by numbers. Lancet Oncol 2005;6:112-117.
resonance imaging. Mol Cancer Ther 2003;2:581-587. 39. Ryu YJ, et al. Glioma: Application of whole-tumor texture analysis of
19. Hamstra DA, et al. Functional diffusion map as an early imaging diffusion-weighted imaging for the evaluation of tumor heterogeneity.
biomarker for high-grade glioma: Correlation with conventional radio- PLoS One 2014;9:e108335.
logic response and overall survival. J Clin Oncol 2008;26:3387-3394. 40. Hamstra DA, et al. Evaluation of the functional diffusion map as an
20. Mardor Y, et al. Early detection of response to radiation therapy in early biomarker of time-to-progression and overall survival in high-
patients with brain malignancies using conventional and high b-value grade glioma. Proc Natl Acad Sci U S A 2005;102:16759-16764.
diffusion-weighted magnetic resonance imaging. J Clin Oncol 2003; 41. Lambin P, et al. Radiomics: Extracting more information from medical
21:1094-1100. images using advanced feature analysis. Eur J Cancer 2012;48:441-
21. Lemasson B, et al. Diffusion-weighted MRI as a biomarker of tumor 446.
radiation treatment response heterogeneity: A comparative study of 42. Lemberskiy G, et al. Time-dependent diffusion in prostate cancer.
whole-volume histogram analysis versus voxel-based functional Invest Radiol 2017;52:405-411.
diffusion map analysis. Transl Oncol 2013;6:554-561. 43. Bongers A, et al. Oscillating gradient diffusion MRI in the ex-vivo
22. Galbán S, et al. DW-MRI as a biomarker to compare therapeutic prostate to assess ADC behavior at small diffusion times, in Pro-
outcomes in radiotherapy regimens incorporating temozolomide or ceedings of the Joint Annual Meeting of the International Society for
gemcitabine in glioblastoma. PLoS One 2012;7:e35857. Magnetic Resonance in Medicine (ISMRM) and ESMRMB. Milan,
23. Pramanik PP, et al. Hypercellularity components of glioblastoma Italy: International Society for Magnetic Resonance in Medicine 2300
identified by high b-value diffusion-weighted imaging. Int J Radiat Clayton Road, Suite 620. Concord, CA, 94520 USA; 2014.
Oncol Biol Phys 2015;92:811-819. 44. Van AT, Holdsworth SJ, Bammer R. In vivo investigation of restricted
24. Panagiotaki E, et al. Noninvasive quantification of solid tumor micro- diffusion in the human brain with optimized oscillating diffusion
structure using VERDICT MRI. Cancer Res 2014;74:1902-1912. gradient encoding. Magn Reson Med 2014;71:83-94.
25. Does MD, Parsons EC, Gore JC. Oscillating gradient measurements of 45. Baron CA, Beaulieu C. Oscillating gradient spin-echo (OGSE)
water diffusion in normal and globally ischemic rat brain. Magn Reson diffusion tensor imaging of the human brain. Magn Reson Med 2014;
Med 2003;49:206-215. 72:726-736.

View publication stats