Key words MR-diffusion/perfusion - liver tumors - technical aspects
Purpose
Quantification of the apparent diffusion coefficient (ADC) from diffusion-weighted
MR imaging (DWI) is widely used both for the evaluation of diffuse liver disease as
well as for the characterization and assessment of treatment response of focal liver
lesions [1 ]
[2 ]
[3 ]
[4 ]
[5 ]. The ADC can provide information on alterations of tissue cellularity, extracellular
space tortuosity, and integrity of cell membranes (e. g. in developing necrosis) [1 ]
[2 ]
[3 ]
[4 ]
[5 ]. Conventionally, the ADC is determined from diffusion-weighted images acquired with
b-values between 0 and 500 – 1000 s/mm2 assuming a mono-exponential relationship between signal intensity and the b-value
[1 ]
[3 ]. However, DWI is not only sensitive to molecular diffusion, but also to other intravoxel
incoherent motion (IVIM) like perfusion due to pseudorandom organization of the capillary
network at the voxel level [4 ]
[5 ]
[6 ]. Perfusion leads to additional signal attenuation at low b-values, whereas signal
attenuation at higher b-values is mainly caused by molecular diffusion (true diffusion
coefficient D) [7 ]
[8 ]. A refined analysis can be performed based on the IVIM theory and the acquisition
of at least four b-values. Influences of diffusion and perfusion can be separated
assuming a bi-exponential behavior of signal intensity, yielding the diffusion coefficient
D, the pseudo-diffusion coefficient D* and the perfusion fraction f [6 ]
[8 ]. In this model D represents the mobility of water molecules in tissue and depends
on the cellularity, tortuosity of the extracellular space, integrity of cell membranes,
and viscosity of fluids [4 ]
[6 ]
[8 ]. f reflects the relative contribution of microvascular blood flow to the DWI signal,
and D* depends on the blood velocity and length of microvessel segments [6 ]
[8 ].
In current research, IVIM model-based analysis of DWI data is employed in diffuse
[9 ]
[10 ]
[11 ]
[12 ]
[13 ] and focal liver disease [14 ]
[15 ]
[16 ]
[17 ]
[18 ]. However, in malignant liver lesions an IVIM analysis based on conventionally used
bi-exponential fitting is difficult due to low D*- and f-values leading to fitting
failures and poor reproducibility, especially for perfusion-related parameters [14 ]
[16 ]
[19 ]
[20 ]
[21 ]. Thus, for oncological liver applications, a simplified IVIM analysis based on only
3 b-values as introduced by Le Bihan [6 ] was recently used in several studies, yielding a stable voxel-wise estimation of
D and f (called D' and f') [22 ]
[23 ]
[24 ]
[25 ]
[26 ]. Initial experience with this approach is promising, but data regarding measurement
repeatability and reproducibility are lacking. In this context “repeatability” informs
on measurement variations due to technically inconsistent readings by the same operator
obtained on the identical subject with the same measurement and analysis equipment
in a short time interval. It therefore provides information on the minimal achievable
measurement error by controlling external sources of errors. “Reproducibility” on
the other hand includes additional potential sources of variation such as a longer
time interval between examinations (e. g. several days) [4 ].
The purpose of this study was to evaluate the repeatability of this simplified 3 b-value
IVIM analysis method, excluding possible influencing factors such as patient positioning
within the MR scanner or confounding long-term variation in liver perfusion.
Materials and Methods
Patients
24 patients (16 male, 8 female, mean age: 67 years [range: 49 – 86]) with hepatic
malignancies underwent 29 standard liver MRI examinations with repeated acquisition
of DWI in the clinical routine between October 2015 and March 2016. Data analysis
was approved by the local institutional review board and patients gave written consent.
Patients suffered from the following primary or secondary malignancies: hepatocellular
cancer (n = 10), neuroendocrine tumor (n = 5), renal cell cancer (n = 3), adenoid
cancer (n = 2), colorectal cancer (n = 1), ovarian cancer (n = 1), uveal melanoma
(n = 1), cholangiocellular carcinoma (n = 1). The majority of patients (n = 16) had
undergone radioembolization, two patients had received chemoembolization, two were
treated by chemotherapy, while four patients were treatment naive. In 12 cases tumor
manifestation was unilobar, while the remaining 12 patients suffered from bilobar
disease. 11 patients presented with liver cirrhosis. None of the patients suffered
from hemochromatosis.
MRI Technique
All patients underwent MR imaging examinations of the liver on the same clinical 1.5 T
MRI scanner in supine position (Philips Healthcare, Best, The Netherlands; Ingenia;
gradient system: maximum amplitude of 45 mT/m, maximum slew rate of 200 T/m/s). A
commercially available phased-array surface coil was used for signal reception. The
standardized imaging protocol comprised a respiratory-triggered single-shot spin-echo
echo-planar DWI sequence ([Table 1 ]) with motion-probing gradients in three orthogonal directions and 3 b-values (0,
50, 800 s/mm2 ). The DWI sequence was acquired twice prior to contrast agent injection without moving
the patient within the MR scanner. Isotropic diffusion-weighted images were reconstructed
on the MRI system.
Table 1
Diffusion-weighted imaging (DWI) sequence parameters.
Tab. 1 Sequenzparameter der diffusionsgewichteten Sequenz (DWI).
FOV (RLxAP)/orientation
380 × 326 mm/transverse
slice number/thickness/gap
28/7.0 mm/0.7 mm
matrix/resolution
112 × 94/3.4 × 3.5 mm
echo time (TE)
63 ms
repetition time (TR)
1 respiratory cycle
imaging time per respiration
1600 ms
EPI-/half-Fourier-/SENSE-factor
51/0.6/2
diffusion gradients
3 orthogonal directions
b-values
0 and 50 s/mm2 (NSA = 2), 800 s/mm2 (NSA = 6)
fat suppression method
SPIR
water-fat shift/BW
9.2 Pixel/23.6 Hz
BW in EPI frequency direction
1437.9 Hz
acquisition time
around 3 min with respiratory gating
SENSE: parallel imaging with sensitivity encoding, FOV: field of view, RL: right–left,
AP: anterior–posterior, EPI: echo-planar imaging, NSA: number of averages, SPIR: spectral
presaturation by inversion recovery, BW: bandwidth. SENSE: Parallele Bildgebung FOV: Sichtfeld, RL: rechts-links, AP: anterior-posterior,
EPI: Echo-planare Bildgebung, NSA: Anzahl der Mittelungen, SPIR: Spektrale Fettsättigung,
BW: Bandbreite
Image Analysis and Definitions
Image analyses were performed in consensus by a radiologist with 5 years of abdominal
imaging experience, and a physicist with more than 17 years of experience in DWI.
Both were blinded to patient-related information.
According to the IVIM concept of Le Bihan et al. [6 ], a two-compartment model of extravascular and intravascular space was applied:
with D true diffusion coefficient, D* pseudodiffusion coefficient, f perfusion fraction,
and S(b) and S(0) signal intensities with and without motion probing gradients, respectively.
By using the following simplified approach (high b-value approximation) as originally
introduced by Le Bihan [6 ]:
D and f of the IVIM model were estimated as D' and f' as recently applied to abdominal
imaging for b-values b0 = 0, b1 = 50 and b2 = 800 s/mm2
[22 ]
[23 ]
[24 ]
[25 ]:
S(b) and S(0) are the signal intensities with and without motion-probing gradients.
The apparent diffusion coefficient ADC(0,800) was also calculated:
Parameter maps were generated by voxel-wise calculation of ADC(0,800), D' and f' with
dedicated software written in MATLAB (Math Works, Natick, MA).
Both DWI scans of the patient were displayed simultaneously. In each patient a target
liver lesion with a diameter of ≥ 1 cm was selected per liver lobe (if present). For
IVIM analysis of the first acquired DWI dataset a hand-drawn region of interest (ROI)
was placed within a central slice of each lesion, avoiding (if possible) noticeable
motion artifacts, pixel misalignments and susceptibility artifacts. Areas close to
the rim of the lesion were excluded to avoid partial volume effects. A second ROI
was placed in an adjacent area of tumor-free liver parenchyma. Large blood vessels
were avoided. All ROIs were drawn on b = 800 s/mm2 images. ROI positions were visually cross-checked between all diffusion-weighted
images and then copied into the parameter maps. For the second DWI dataset, the same
shape and size of the ROIs was used, but ROI positions were adapted to the actual
lesion/liver position which may vary slightly between scans. For each ROI, the mean
parameter value and standard deviation were determined and compared between the first
and second measurement.
Measurement repeatability of the ADC(0,800), D' and f' for liver lesions and parenchyma
was assessed by calculating the intra-session coefficient of variation (CV) and Lin’s
concordance correlation coefficient (CCC). Measurement repeatability was rated according
to the intra-session CV as excellent (CV ≤ 10 %), good (CV between 10 and 20 %), acceptable
(CV between 20 % and 30 %) and poor (CV > 30 %) [14 ]. The grade of agreement between the two measurements was defined according to the
CCC as almost perfect (CCC > 0.90), substantial (CCC between 0.80 and 0.90), moderate
(CCC between 0.65 and 0.80) and poor (CCC < 0.65) [27 ]. Furthermore, we compared the CVs for lesions and liver parenchyma, right and left
liver lobes (in patients with bilobar disease), locally treated and untreated lesions
as well as cirrhotic and non-cirrhotic liver parenchyma.
Statistical Analysis
Statistics were performed using commercially available software (SPSS, version 22.0,
IBM, Armonk, NY). Normal distribution of the data was assessed using Q-Q-plots. Bland-Altman
analysis was performed to compare the mean relative differences of the parameters
normalized to the mean values of the two measurements. Statistical significance (p < 0.05)
for group differences was tested with Student’s t-test for independent samples in
case of intersubject comparisons, and with Student’s t-test for paired samples in
case of intrasubject comparisons.
Results
Overall 86 ROIs (43 in liver lesions, 43 in liver parenchyma) were analyzed. The results
of statistical analyses are summarized in [Table 2 ].
Table 2
Measurement results and test-retest repeatability of ADC, D' and f' for malignant
liver lesions and liver parenchyma.
Tab. 2 Messergebnisse und Test-Retest-Repeatability von ADC, D' und f' von malignen Leberläsionen
und Lebergewebe.
malignant lesions
liver parenchyma
overall (n = 43)
right (n = 23)
left (n = 20)
overall (n = 43)
right (n = 23)
left (n = 20)
ADC (0,800)
test
1417 ± 412 (529 – 2864)
1514 ± 428 (981 – 2864)
1298 ± 369 (529 – 2060)
1359 ± 205 (968 – 1925)
1394 ± 233 (1036 – 1925)
1319 ± 165 (968 – 1657)
retest
1418 ± 399 (637 – 2698)
1480 ± 442 (935 – 2698)
1347 ± 341 (637 – 2002)
1363 ± 216 (1020 – 1888)
1387 ± 235 (1055 – 1888)
1336 ± 195 (1020 – 1729)
mean
1416 ± 395 (588 – 2781)
1497 ± 426 (980 – 2781)
1322 ± 342 (588 – 1904)
1361 ± 194 (1032 – 1895)
1390 ± 224 (1123 – 1895)
1328 ± 150 (1032 – 1601)
wSD
103 ± 81 (2 – 271)
108 ± 80 (2 – 271)
103 ± 87 (5 – 249)
98 ± 64 (7 – 247)
84 ± 44 (7 – 165)
114 ± 80 (17 – 247)
CV [%]
7.5 ± 5.8 (0.2 – 19.1)
6.9 ± 5.1 (0.2 – 16.4)
8.1 ± 6.7 (0.3 – 19.1)
7.3 ± 4.8 (0.4 – 18.8)
6.4 ± 3.8 (0.4 ± 13.6)
8.4 ± 5.7 (1.4 – 18.8)
mean percentage Δ (95 % LoA) [%]
0.14 (– 24.77 – 25.91)
– 2.21 (– 26.34 – 20.76)
3.73 (– 23.72 – 32.60)
0.29 (– 24.28 – 24.76)
– 0.48 (– 21.52 – 20.46)
1.35 (– 27.41 – 29.67)
CCC
0.894
0.910
0.853
0.688
0.832
0.388
true diffusion coefficient D'
test
1297 ± 396 (503 – 2807)
1416 ± 433 (898 – 2807)
1160 ± 302 (503 – 1744)
1095 ± 177 (798 – 1491)
1118 ± 176 (799 – 1491)
1069 ± 180 (812 – 1486)
retest
1311 ± 390 (589 – 2643)
1396 ± 444 (832 – 2643)
1212 ± 298 (589 – 2036)
1105 ± 200 (676 – 1602)
1118 ± 206 (731 – 1602)
1091 ± 196 (676 – 1460)
mean
1303 ± 381 (565 – 2725)
1406 ± 430 (898 – 2725)
1186 ± 283 (565 – 1890)
1100 ± 164 (799 – 1536)
1118 ± 174 (864 – 1536)
1080 ± 153 (799 – 1363)
wSD
106 ± 79 (0.1 – 256)
108 ± 74 (0.1 – 244)
110 ± 90 (0.4 – 256)
106 ± 79 (0.7 – 348)
93 ± 72 (0.7 – 206)
121 ± 95 (2 – 348)
CV [%]
8.5 ± 6.4 (0.01 – 23.0)
7.5 ± 5.5 (0.01 – 19.6)
9.5 ± 7.4 (0.03 – 23.0)
9.8 ± 7.3 (0.1 – 28.1)
8.8 ± 6.5 (0.1 – 21.7)
10.9 ± 8.2 (0.2 – 28.1)
mean percentage Δ (95 % LoA) [%]
0.93 (– 28.43 – 30.77)
– 1.39 (– 27.87 – 23.91)
4.35 (– 27.80 – 32.36)
0.91 (– 33.50 – 34.70)
– 0.01 (– 31.25 – 30.33)
2.10 (– 36.43 – 40.05)
CCC
0.886
0.919
0.772
0.501
0.646
0.319
perfusion fraction f'
test
96.0 ± 75.7 (3.5 – 310)
85.7 ± 70.8 (3.5 – 310)
108 ± 81.1 (16.4 – 306.8)
181.5 ± 74.5 (68.8 – 374.4)
187 ± 81 (68.8 – 374.4)
175.4 ± 67.9 (93.0 – 342.8)
retest
99.2 ± 72.7 (1.7 – 273.3)
83.5 ± 58.8 (1.7 – 233)
117.2 ± 84.0 (15.1 – 273.3)
177.2 ± 78.8 (76.7 – 459.6)
180.3 ± 83.8 (76.7 – 459.6)
173.6 ± 74.7 (80.6 – 377.9)
mean
97.6 ± 73.1 (2.6 – 290.1)
84.6 ± 64.3 (2.6 – 267.1)
112.5 ± 81.2 (15.8 – 290.1)
179.4 ± 74.2 (72.8 – 412.0)
183.6 ± 80.5 (72.8 – 412.0)
174.5 ± 68.1 (92.2 – 360.4)
wSD
10.5 ± 14.8 (0.1 – 60.9)
7.7 ± 12.0 (0.1 – 60.9)
13.9 ± 17.0 (0.1 – 59.5)
22.3 ± 15.7 (1.1 – 67.4)
20.0 ± 14.9 (1.9 – 67.4)
25.0 ± 16.5 (1.1 – 64.8)
CV [%]
11.0 ± 11.4 (0.1 – 49.0)
9.5 ± 10.2 (0.1 – 49.0)
12.7 ± 12.6 (0.2 – 47.8)
13.0 ± 9.2 (1.0 – 36.0)
10.5 ± 5.3 (1.0 – 20.1)
16.0 ± 11.7 (1.2 – 36.0)
mean percentage Δ (95 % LoA) [%]
2.51 (– 40.34 – 46.80)
– 2.54 (– 39.31 – 38.72)
8.26 (– 40.77 – 55.34)
– 2.40 (– 47.10 – 41.03)
– 3.52 (– 35.61 – 29.23)
– 1.07 (– 58.45 – 52.57)
CCC
0.939
0.950
0.928
0.872
0.906
0.818
wSD: within subject standard deviation, CV: intra-session coefficient of variation,
CCC: concordance correlation coefficient, 95 %-LoA: Bland-Altman limits of agreement. wSD: Standardabweichung innerhalb des Patienten, CV: Variationskoeffizient, CCC: Konkordanzkorrelationskoeffizient,
95 %-LoA: Übereinstimmungsgrenzen der Bland-Altman Analyse.
Comparing ADC(0,800), D' and f' values between test and retest measurements, no significant
differences were found, neither for liver lesions nor for the parenchyma. Bland-Altman
analysis revealed no systematic error between both measurements ([Table 2, ]
[Fig. 1 ]). Example images and parameter maps are shown in [Fig. 2 ].
Fig. 1 Bland-Altman plots for the conventional ADC(0,800), the diffusion sensitive parameter
D' and the perfusion sensitive parameter f' of lesions in the left a and right b liver lobe, as well as for liver parenchyma in the left c and right d liver lobe.
Abb. 1 Bland-Altman-Diagramme für den konventionellen ADC(0,800), den diffusionssensitiven
Parameter D' und den perfusionssensitiven Parameter f' von Läsionen des linken a und rechten b Leberlappens sowie für Leberparenchym des linken c und rechten d Leberlappens.
Fig. 2 Typical example of intravoxel incoherent motion (IVIM)-based parameter maps for a
multifocal hepatocellular carcinoma (HCC). Original diffusion-weighted images with
b = 0, 50, 800 s/mm2 are presented together with conventional ADC(0,800), diffusion sensitive D' and perfusion
sensitive f' parameter maps for a initial and b repeated acquisition. The parameter maps are displayed as color-coded overlays over
DWI b = 0. Analyzed regions of interest are marked in white. For ADC(0,800), D' and
f', a coefficient of variation (CV) of 0.08 %, 1.61 % and 12.29 %, respectively, was
reached.
Abb. 2 Typisches Beispiel für Intravoxel-Incoherent-Motion (IVIM) basierte Parameterkarten
eines Patienten mit multifokalem hepatozellulärem Karzinom (HCC). Original diffusionsgewichtete
Bilder mit b = 0,50, 800 s/mm2 sind zusammen mit dem konventionellen ADC (0,800), den diffusionssensitiven D'- und
den perfusionssensitiven f'-Parameterkarten für die a erste und b zweite Akquisition abgebildet. Die Parameterkarten sind als farbkodiertes Overlay
über dem DWI-Bild mit b = 0 dargestellt. Die analysierten Regions-of-Interest sind
in weiß eingezeichnet. Für ADC(0,800), D' und f' ergaben sich Variationskoeffizienten
von 0,08 %, 1,61 % und 12,29 %.
Overall, measurement repeatability of the ADC(0,800) and D' was excellent both for
the liver parenchyma and lesions [ADC(0,800): CV 7.3 % and 7.5 %, respectively; D':
CV 9.8 % and 8.5 %, respectively], while repeatability of f' was good (CV 13.0 % and
11.0 %).
While the repeatability of the ADC(0,800) was comparable between lesions and parenchyma
(p = 0.899), the CVs of D' and f' were slightly higher for the liver parenchyma compared
to lesions, but without reaching statistical significance (p = 0.339 and p = 0.353,
respectively). When only the right liver lobes were included, which are less susceptible
to motion artifacts, there were also no significant differences between CVs of lesions
and the parenchyma (p = 0.702 for the ADC(0,800), p = 0.466 for D' and p = 0.690 for
f').
Overall, the parenchyma and lesions in the right liver lobe showed better repeatability
of all parameter values compared to the left liver lobe. In patients presenting with
bilobar disease, a non-significant trend was found for ADC, D' and f' towards higher
CVs for the left liver lobe compared to the right lobe, both for the liver parenchyma
(p = 0.097, p = 0.191 and p = 0.158, respectively) and lesions (p = 0.469, p = 0.282
and p = 0.748, respectively). The best repeatability was observed for the ADC(0,800)
of the parenchyma and lesions in the right liver lobe (6.4 % and 6.9 %); worst values
for f' in the left liver lobe (16.0 % and 12.7 %).
There was substantial to almost perfect correlation between the two measurements for
the ADC(0,800), D' and f' of liver lesions with slightly lower correlation coefficients
for lesions in the left liver lobe ([Table 2 ]). For the liver parenchyma there was only poor to moderate correlation between the
measurements for the ADC(0,800) and D' with poor values especially occurring in the
left liver lobe. In contrast, there was a substantial inter-measurement correlation
for f'-values with almost perfect correlation in the right liver lobe.
Additional comparison of measurement repeatability between clinically defined groups
yielded the following results: Compared to locally treated lesions (n = 29), untreated
lesions (n = 6) showed significantly higher CVs for the ADC and D' (p = 0.006 and
p = 0.014, respectively), whereas the CVs of f'-values did not differ significantly
(p = 0.385). When comparing cirrhotic and non-cirrhotic liver parenchyma, the CVs
did not differ significantly for ADC(0,800), D' and f' (p = 0.389, p = 0.528 and p = 0.247,
respectively).
Discussion
Intravoxel incoherent motion (IVIM) model-based analysis can yield information on
the amount of microvasculature (perfusion fraction f), blood-flow velocity and vessel
architecture (pseudo-diffusion coefficient D*), and is increasingly being investigated
for application in body imaging [6 ]
[8 ]. However, before wider clinical application of this technique can be recommended,
technical parameters especially the repeatability and reproducibility of measurements
need to be determined. Although quantitative IVIM-based analysis of DWI data in the
liver is increasingly being employed both for diffuse [9 ]
[10 ]
[11 ]
[12 ]
[13 ] and focal liver disease [14 ]
[15 ]
[16 ]
[17 ]
[18 ], most data on measurement reproducibility to date have been obtained for normal
liver tissue [16 ]
[28 ]
[29 ]
[30 ].
For bi-exponential IVIM analysis mostly unconstrained non-linear least squares (NLLS)
fitting procedures with simultaneous determination of D, D* and f (so-called full
fitting) are used. However, these are problematic in malignant liver lesions. Weak
bi-exponential signal decay (low D*) and weak IVIM effect (low f) as observed in malignant
lesions lead to fitting failures and poor reproducibility [14 ]
[16 ]
[19 ]
[20 ]
[21 ]. Improved stability can be achieved for IVIM approaches using a high b-value approximation
of a mono-exponential signal for D determination like segmented fitting [31 ]
[32 ]
[33 ] and simplified IVIM [34 ], which both decrease the degree of freedom by determining the parameters step-by-step. In
segmented fitting, parameters are determined by fitting procedures, whereas in simplified
IVIM explicit approximation formulas in combination with three or four acquired b-values
are used. D and f can be approximated from only three b-values and D, f and D* from
only four b-values. For lesion characterization, it turned out that simplified IVIM
with four b-values had no further benefit to simplified IVIM with three b-values [34 ]. By using simplified IVIM based on three b-values, a stable voxel-wise estimation
of the diffusion coefficient (D') and the perfusion fraction (f') within acceptable
acquisition times can be performed [22 ]
[23 ]
[24 ]
[25 ]
[26 ].
The aim of the present study was to determine the measurement repeatability of the
three b-value approach, i. e. consistency of measurement results while minimizing
errors due to long-term biological variability [4 ]. We chose to determine repeatability instead of reproducibility because reproducibility
can additionally be influenced by long-term biological changes for example in portal
venous flow [35 ]. A recent study showed no significant differences between the repeatability and
reproducibility of IVIM parameters determined by NLLS fitting for the liver parenchyma
[28 ]. This means that variations of the NLLS fitting results were greater than changes
due to repositioning of the patients or physiological long-term changes. This might
be different for simplified IVIM.
With regard to repeatability and reproducibility, a comparison of different IVIM-based
studies is difficult as measurement results depend on the choice of b-values, field
strength, acquisition technique, methodology of parameter analysis, and the investigated
target. The reported reproducibility of overall ADC and D was generally better than
that of f and D* irrespective of the analysis method but showed a wide range of variation:
for the ADC and D, the coefficients of variation (CV) ranged from 2.3 – 15.6 % and
3.2 – 25.3 %, respectively, while the maximum percentage variation of f and D* has
been described as high as 7.7 – 241 % and 14.6 – 2120 %, respectively [9 ]
[14 ]
[16 ]
[28 ]
[29 ]
[30 ]
[36 ]
[37 ]
[38 ] ([Table 3 ]). Influencing factors of repeatability and reproducibility leading to this wide
range of variation between different studies are discussed in the following paragraphs:
Table 3
Summary of study results.
Tab. 3 Zusammenfassung der Ergebnisse publizierter Studien.
reference
field strength
b-values in s/mm2
motion control
analysis
investigated tissue
parameter
ADC
D
f
D*
Gurney-Champion et al. 2016 [28 ]
3.0 T
0, 10, 20, 30, 40, 50, 65, 80, 100, 125, 175, 250, 375, 500
RT
NLLS biexp fit, ROI-wise
liver healthy
large ROI (n = 16)
small ROI (n = 16)
CV
NA
NA
12 %
13 %
47 %
34 %
150 %
77 %
Lee et al. 2015 [29 ]
1.5 T
0, 30, 60, 100, 150, 200, 400, 600, 900
RT/FB/ET
NLLS biexp fit, voxel-wise, right and left lobe
liver healthy
RT, right lobe (n = 12)
FB, right lobe (n = 12)
CV
2.3 %
4.3 %
3.2 %
4.1 %
11.9 %
15.8 %
75.0 %
74.2 %
Andreou et al. 2013 [16 ]
1.5 T
0, 25, 50, 75, 100, 250, 500, 900
RT
NLLS biexp fit, voxel-wise
liversurr. (n = 14)
Met. (n = 14)
max. Δ [%]
6.8 %
14.7 %
8.1 %
25.3 %
25.1 %
241 %
59 %
2120 %
Dyvorne et al. 2014 [30 ]
1.5 T
0, 15, 30, 45, 60, 75, 90, 105, 120, 135, 150, 175, 200, 400, 600, 800
RT
NLLS biexpfit, ROI-wise, right lobe, bipolar DW-gradients
liver healthy and diffuse disease (n = 14)
CV
NA
12.0 %
32.3 %
193.8 %
Dyvorne et al. 2013 [37 ]
1.5 T
0, 15, 30, 45, 60, 75, 90, 105, 120, 135, 150, 175, 200, 400, 600, 800
RT
FB
Bayesian fit, ROI-wise, right lobe, bipolar DW-gradients
liver healthy (n = 20)
diffuse disease (n = 20)
CV
NA
NA
9.3 %
6.8 %
17.9 %
35.6 %
37.3 %
55.5 %
Jajamovich et al. 2014 [36 ]
3.0 T
0, 15, 30, 45, 60, 75, 90, 105, 120, 135, 150, 175, 200, 400, 600, 800
FB
Bayesian fit, ROI-wise, right lobe
liver healthy and diffuse disease (n = 30)
CV
8.2 %
15.2 %
20.4 %
51.6 %
Kakite et al. 2015 [14 ]
3.0 T
0, 15, 30, 45, 60, 75, 90, 105, 120, 135, 150, 175, 200, 400, 600, 800
FB
Bayesian fit, ROI-wise, right lobe in case of liver
liversurr. (n = 11)
HCC (n = 15)
HCC right (n = 10)
HCC left (n = 5)
CV
8.8 %
15.6 %
11.6 %
18.4 %
13.2 %
19.7 %
19.0 %
6.8 %
25.3 %
37.3 %
32.9 %
57.0 %
59.0 %
60.6 %
61.4 %
53.1 %
Cohen et al. 2015 [9 ]
1.5 T
0, 10, 25, 50, 100, 150, 200, 400, 800
RT
segmented fitting
liver healthy
voxel-wise
ROI-wise
CV
10 %
28 %
Patel et al. 2010 [38 ]
1.5 T
0,50, 100, 150, 200, 300, 500, 700, 1000
FB/RT
segmented fitting, ROI-wise, only right lobe
liver healty
RT (n = 5)
FB (n = 4)
CV
3.0 %
3.4 %
3.8 %
6.4 %
7.7 %
16.0 %
14.6 %
39.1 %
Current study (Pieper et al.)
1.5 T
0, 50, 800
RT
simplified approach, voxel-wise
liver surr. (n = 24)
malignant (n = 24)
CV
7.3 %
7.5 %
9.8 %
8.5 %
13.0 %
11.0 %
NA
NA
T: Tesla; FB: free-breathing; RT: respiratory triggering or navigation; CV: coefficient
of variation. CVs of published studies represent reproducibility, whereas CVs in the
present study represent repeatability. T: Tesla; FB: freie Atmung; RT: Atemtriggerung oder -navigation; CV: Variationskoeffizient.
Die CVs der publizierten Studien geben Reproducibility wieder, während die CVs der
aktuellen Studie die Repeatability wiedergeben.
First, measurement reproducibility especially of perfusion-associated parameters depends
on the method of data analysis. This is especially true for malignant lesions. Full
NLLS fitting yielded the worst reproducibility of IVIM parameters especially in malignant
lesions [16 ]
[28 ]
[29 ]
[30 ]. In a study employing voxel-wise NLLS (with respiratory triggering), the reproducibility
of f and D* was considerably worse for metastases compared to the surrounding liver
parenchyma [16 ]. By comparison, Bayesian fitting yielded better reproducibility than NLLS fitting
with only slightly worse results in malignant lesions compared to normal liver [14 ]
[36 ]
[37 ]. Segmented fitting produced the lowest variation coefficients of IVIM parameters
but has not yet been evaluated for malignant lesions [9 ]
[10 ]
[11 ]
[12 ]
[13 ]
[38 ]. For our presented simplified IVIM approach with voxel-wise analysis (respiratory
triggering), the CVs (repeatability) of D' and f' in the case of normal liver (9.8 %
and 13.0 %, respectively) were comparable to the CVs (reproducibility) of other analysis
methods. However, the coefficients of variation of D' and f' of malignant lesions
were better than those obtained with other methods to date. However, the reproducibility
needs to be determined for the simplified approach, especially in focal lesion changes
due to repositioning. Overall, similar or even better repeatability was reached in
malignant lesions compared to surrounding tissue (8.5 vs. 9.8 % for D' and 11.0 %
vs. 13.0 % for f'). This may be due to differences between the methods of parameter
calculation/fitting. In contrast to fitting-based approaches, the simplified method
is not limited by low D* and f values. Results depend primarily on the signal-to-noise
ratio, which is typically higher in malignant lesions due to stronger diffusion restriction,
lower perfusion effects and longer T2-relaxation times [39 ].
Second, measurement reproducibility is influenced by the method of data acquisition.
Free-breathing DWI acquisition is associated with pixel misalignments which may lead
to measurement errors especially when examining circumscribed lesions, like malignant
tumors. The use of respiratory triggering as performed in our work can improve reproducibility.
As this has only been done before for liver tissue and not for malignant lesions [29 ]
[37 ]
[38 ], we cannot put our data into perspective.
Third, repeatability is influenced by physiological processes. As observed previously,
the repeatability of IVIM parameters in our study for left hepatic lobes was worse
compared to that for right lobes, probably due to cardiac motion. Repeatability may
further be improved by echocardiography triggering [29 ]. However, combined cardiac and respiratory triggering leads to a considerable penalty
in acquisition time and is therefore not routinely employed.
Fourth, prior treatment may also influence repeatability. Even though no statistically
significant differences have been described for treated and untreated HCCs for ROI-wise
Bayesian fitting [14 ], our study found better repeatability of ADC and D' for liver lesions after treatment
by radioembolization or chemoembolization. These therapies usually induce a certain
degree of tissue necrosis within the lesion which is associated with increasing ADC
values [1 ]
[2 ]
[3 ]
[5 ]. This in turn is associated with an increased signal-to-noise ratio and constant
absolute measurement errors, so post-therapeutic relative measurement accuracy will
be higher.
For clinical application it is important to detect differences in IVIM parameters
between different groups of patients or changes after treatment which have to be larger
than the measurement error. Thus, adequate measurement reproducibility of the quantified
parameters is essential. In this study, we demonstrated good repeatability values
of the simplified IVIM approach, which explains the successful use in first clinical
oncological liver applications. The simplified IVIM approach was able to improve lesion
characterization [25 ], allowed for identification of patients likely to develop early blood-flow stasis
during resin-based radioembolization [23 ], and improved response assessment after radioembolization [24 ]. Patients developing early stasis were characterized by a significantly lower estimated
perfusion fraction f' by a mean of about 60 % [23 ]. Likewise patients showing tumor growth or shrinkage after treatment showed relative
group differences of f' and D' of about 50 % [24 ]. In both studies group differences are considerably above the measurement variation
observed in the present study.
The results of our study are limited by several factors. First, we included different
tumor types and treated and untreated lesions in order to investigate the overall
effect over a wide spectrum of existing parameter values. A separate analysis of different
tumor entities would also be interesting but needs larger patient numbers. Second,
measurement repeatability instead of reproducibility was examined to avoid biological
alterations. However, the influence of repositioning should also be investigated in
the future for simplified IVIM. When using NLLS fitting, no statistically significant
differences were found between repeatability and reproducibility for liver tissue
which means that the error due to repositioning is smaller than the error of the fitting
procedure [28 ]. This might be different for simplified IVIM and in case of lesions. This is particularly
important when employing IVIM analysis for follow-up after therapy in which typically
the second examination is performed several weeks later. Third, as described above,
we investigated simplified IVIM with three b-values. Thus, the determination of D*
was not possible. However, it provides numerically stable voxel-wise calculation of
parameter maps of D' and f' at low acquisition times. These maps can be clinically
used for visual inspection and D' and f' may serve as standardized empirical biomarkers
indicating nonspecific pathological changes or therapy responses. Test and retest
measurements with four b-value simplified IVIM may be interesting to investigate in
further studies. As in other IVIM-based procedures, analysis remains challenging due
to imaging artifacts in DWI. Thus, experienced operators are needed to recognize artifacts
that may corrupt measurement results.
Conclusion
In conclusion, the present study revealed excellent measurement repeatability of the
ADC(0,800) and the estimated diffusion coefficient D', as well as good repeatability
of the estimated perfusion fraction f' calculated from only three b-values using the
simplified IVIM approach. In contrast to non-linear bi-exponential fitting-based approaches,
the simplified approach yielded satisfactory results not only in liver tissue but
also in malignant liver lesions in which fitting is difficult. Thus, the simplified
approach may be helpful in the diagnosis and therapy monitoring of liver malignancies.