Internal comparison standard for abdominal diffusion-weighted imaging

Abstract

Background

Standards for abdominal diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC) measurements, and analysis are required for reproducibility.

Purpose

To identify optimal internal comparison standards for DWI to normalize the measured ADC for increased accuracy of differentiating malignant and benign abdominal lesions.

Material and Methods

We retrospectively studied 97 lesions (89 patients; age, 57 ± 13 years) with histopathologically confirmed abdominal disease. Seven normal body parts/contents (normal parenchyma, spleen, kidney, gallbladder bile, paraspinal muscle, spinal cord, and cerebrospinal fluid [CSF]) were assessed as internal references for possible use as comparison standards. Three observers performed ADC measurements. Statistical analyses included interclass correlation coefficients (ICCs), Mann–Whitney and Kruskal–Wallis tests, and coefficient of variation (CV). ROC analyses were performed to assess diagnostic accuracy of lesion ADC and normalized ADC for differentiating lesions. Pathology results were the reference standard.

Results

Mean and normalized ADCs were significantly lower for malignant lesions than for benign lesions (P < 0.001). ICC was excellent for all internal references. Gallbladder had the lowest CV. Receiver operating characteristic (ROC) analyses showed that normalized ADCs obtained using normal parenchyma were better than lesion ADCs for differentiating malignant and benign abdominal lesions (area under the curve [AUC], 0.808 and 0.756, respectively). The normalized ADCs obtained using CSF shows higher accuracy than lesion ADCs (0.80 and 0.76, respectively) for differentiating between malignant and benign abdominal lesions.

Conclusion

The normal parenchyma from a lesion-detected organ can be used as an internal comparison standard for DWI. CSF can be used as a generalizable in plane reference standard.

Keywords

Diffusion-weighted imaging (DWI)magnetic resonance imaging (MRI)internal reference comparison standard normalization abdominal

Introduction

In recent years, diffusion-weighted imaging (DWI) has been accepted as a successful imaging technique and has evolved into a method used worldwide for characterizing malignant and benign abdominal lesions (1,2). The apparent diffusion coefficient (ADC) has been recognized as a cancer-imaging biomarker for diagnosing malignancy and monitoring the early treatment response (3,4). Although the ADC is frequently used to differentiate neoplasms, techniques and imaging parameters change (5). Standards for DWI and ADC measurements, analysis, and display are required for reproducibility. Significant inter-vendor differences have been reported regarding ADC values of abdominal organs (6). Improved phantoms mimicking the cellular environment of living tissue are required (3). Artificially created external ADC phantoms, however, cannot simulate the complex properties of tissues or lesions and the environment in the body (6). Data concerning the internal reference for measuring the ADC, however, are limited (7). Liver, spleen, gallbladder, pancreas, kidneys, and paraspinal muscles have been evaluated in previous studies (6,7) to fill the need for a relatively stable internal reference as a comparison standard when measuring ADCs.

The purpose of this study was to explore the possibilities for identifying an optimum internal comparison standard for DWI to normalize the ADC and increase the accuracy of differentiating malignant and benign abdominal lesions.

Material and Methods

Patients

Our institutional review board approved the research protocol of this retrospective study. The study parallels another, previously reported study (8). Between March 2008 and June 2010, we enrolled 108 consecutive patients (127 lesions) who were referred for magnetic resonance imaging (MRI) because of a possible abdominal lesion. Inclusion and exclusion criteria were summarized on flow chart (Fig. 1). Only lesions that had at least measurable solid components were included for evaluation. The remaining 97 lesions in 89 patients (34 women, 55 men; age, 57 ± 13 years) were included in the study. Eight patients had two lesions and each of the others had one lesion. Within a median of 11 days (range, 0–37 days) after MRI, specimens for pathological evaluation were obtained by biopsy (45 lesions in 44 patients) or during surgery (52 lesions in 45 patients). None of the patients underwent any other treatment or intervention during this time interval. Pathology results were obtained as reference standards. Lesions were categorized as malignant (n = 76) or benign (n = 21) (Table 1).

Fig. 1.

Flow chart shows patients and study population. SCA, sickle cell anemia; immeasurable microscopic lesion, ampulla of Vater adenocancer.

Table 1.

Diagnosis of 97 lesions in 89 patients.

Site	Malignant	n (%)	Benign	n (%)	Total n (%)
Liver and bile duct	Metastases (n = 15), HCC (n = 5), cholangiocellular Ca (n = 5), cholangioadenosquamous Ca (n = 1)	26 (27)	Hemangioma (n = 7), fibrosis (n = 2), cirrhotic nodule (n = 1)	10 (10)	36 (37)
Gallbladder	AdenoCa (n = 2), NHL (n = 1)	3 (3)	−	−	3 (3)
Pancreas– periampullary	Pancreatic ductal adenoCa (n = 11), amp. Vater adenoCa (n = 1), adenosquamous Ca (n = 1)	13 (13)	Groove pancreatitis (n = 1), neuroendocrine tumor (n = 1)	2 (2)	15 (15)
Kidney– urinary	RCC (n = 19), urothelial (n = 3), NHL (n = 2), metastasis (n = 1)	25 (26)	Oncocytoma (n = 1)	1 (1)	26 (27)
Adrenal	Primary adrenal lymphoma (n = 4), adrenal Ca (n = 2)	6 (6)	Adenoma (n = 2), hyperplasia (n = 2), pheochromocytoma (n = 2)	6 (6)	12 (12)
GI system	Stomach GIST (n = 1), stomach Ca (n = 1), colon Ca (n = 1)	3 (3)	Colon leiomyoma (n = 1)	1 (1)	4 (4)
Spleen	–	–	Splenic hemangioma (n = 1)	1 (1)	1 (1)
Total n (%)		76 (78)		21 (22)	97 (100)

denoCa, adenocarcinoma; Ca, carcinoma; GI, gastrointestinal; GIST, gastrointestinal stromal tumor; HCC, hepatocellular carcinoma; NHL, non-Hodgkin’s lymphoma; RCC, renal cell carcinoma.

Imaging protocol

All MRI examinations were performed after 6–8 h of fasting. MRI scans were obtained using a 1.5 T MRI unit (Magnetom Avanto; Siemens, Erlangen, Germany) with a phase-array body coil. In addition to routine abdominal MRI, all patients underwent axial abdominal DWI. These images were acquired using single-shot echo-planar imaging (SSH-EPI) sequences during free breathing. The sequence parameters are summarized in Table 2. Spectral presaturation was applied to suppress chemical shift artifacts. One set of b values (0, 600, 800, and 1000 s/mm²) was used to generate the ADC maps.

Table 2.

Diffusion-weighted imaging sequence parameters.

Sequence	Single shot EPI
Repetition time (ms)	4200–4400
Echo time (ms)	87
Echo spacing (ms)	0.88
Section thickness (mm)	5
Sections (n)	30–35
Section gap (mm)	1 (20%)
Bandwidth (Hz/pixel)	1736
Signals acquired (n)	2
b values (s/mm²)	0, 600, 800, 1000
Field of view (mm)	400 × 300
Matrix	142 × 192
Parallel acceleration factor	GRAPPA-2
EPI factor	156
Fat saturation	Fat saturation

EPI, echo planar imaging; GRAPPA, generalized autocalibrating partially parallel acquisition.

Image analysis

Two radiologists (ZK and GE, who had 14 and 10 years of experience in abdominal MRI, respectively) retrospectively evaluated the images. To evaluate the sampled lesions for pathology on DWI, these two experienced observers reviewed the DWI scans, T2-weighted (T2W) and T1-weighted (T1W) images including in-phase and out-of-phase, and contrast-enhanced images. They arrived at the diagnosis by consensus and the related image numbers were recorded. Later, during different sessions (separated from the lesion-defining process by at least 3 weeks to minimize recall bias), the same two observers randomly analyzed the lesions on the recorded DWI. They were blinded to the patients’ clinical data and any MRI, surgical, and histopathological results.

ADC measurement

The two observers (ZK and GE) measured the lesion ADCs (by consensus) on the measured area of the lesion on ADC maps. The ADC values with standard deviations (SDs) were measured free-hand or as elliptical regions of interest (ROI) in the lesions. In addition to the normal parenchyma of the organ in which a lesion was detected, six other normal body parts or contents—spleen, kidney cortex, gallbladder bile, paraspinal muscles, spinal cord and cerebrospinal fluid (CSF)—were assessed as possible internal references to use as comparison standards. Two observers (GE and EK) measured the ADC values of the normal parenchyma and other internal references independently.

The ROIs were drawn carefully. The ROI was kept as large as possible including solid component without involving the necrotic core of the lesion (Fig. 2a and b), if present, because necrotic portions of lesions do not have characteristic tissue diffusion properties (3). If a lesion was purely cystic, it was entirely included in the ROI. Elliptical ROIs were as large as possible (up to 2 cm²) and were chosen carefully so as to be located in normal parenchyma and, like other internal references, to avoid artifacts and vessels. The ROI size changed according to the lesion size. Specific techniques were used to measure some of the internal references. For example, the upper, clear, homogeneous content of the gallbladder was selected for measurement to avoid the sediment present in some patients (Fig. 2b). Renal cortical measurements were made in the left kidney in all but two cases, in which the right kidney was measured because of left nephrectomy. At least × 2 magnification was used for accurate measurements of small areas, such as spinal cord and CSF (Fig. 2a and b). The relatively broader distal spinal cord was measured (Fig. 2b). We chose to measure the area with the clearest CSF to avoiding overlap with adjacent vascular or neural structure by using broader windowing.

Fig. 2.

Examples of ADC measurements in the lesion and internal references on ADC maps. (a) A 60-year-old man had right renal cell carcinoma. Examples of the lesion, kidney cortex, paraspinal muscle, and spleen measurements. The lesion (open arrows) shows restricted diffusion with a low ADC value. (b) A 48-year-old man had liver hemangioma. Examples of the lesion, normal liver parenchyma, gallbladder, cerebrospinal fluid, and spinal cord measurements. The lesion (open arrow) shows central mild diffusion restriction. The lesion mean ADC value is higher than that of the normal parenchyma of the liver.

The mean coefficient of variation (CV = SD/mean) of the ADC of the internal references and normal parenchyma were calculated, recorded as percentages. Quantitative analysis of the lesions was performed to assess the malignity or benignity by measuring the lesion’s ADC values and normalized ADCs (e.g. lesion ADC/normal parenchyma ADC ratio; lesion ADC/other internal reference ADCs). We were unable to measure these values in 27 internal references from patients who had undergone cholecystectomy (n = 10). Normal parenchyma was not measured because of diffuse liver metastasis (n = 2), cirrhosis (n = 2), or bilateral adrenal lesions (n = 2). If the target organ had no measurable normal parenchyma, any normal parenchyma measurement could not be performed. These organs included the colon (n = 2), extrahepatic bile duct (n = 3), ampulla of Vater (n = 1), and urothelium (n = 3). Liver parenchyma was measured as normal parenchyma for two infiltrative gallbladder mass lesions. The success rates of the internal references were compared.

Statistical analysis

The Wilcoxon signed-rank test was used to compare the ADC values and normalized ADCs. The Mann–Whitney and Kruskal–Wallis tests were used to compare lesion ADC values and normalized ADCs between the malignant and benign groups. The Mann–Whitney test was used for post-hoc analyses if the Kruskal–Wallis test results were significant. Values of P < 0.05 indicated statistical significance. ROC analysis was used to evaluate the diagnostic accuracy of the lesion’s ADC values and normalized ADCs for differentiating between malignant and benign lesions. Comparisons between observer measures of internal references were applied using the Wilcoxon test. Inter-observer agreement was calculated using the reliability test–correlation coefficients. Correlation coefficients were interpreted as indicating a relation that was excellent, r ≥ 0.91; good, 0.90 ≥ r ≥ 0.71; fair 0.70 ≥ r ≥ 0.51; weak 0.50 ≥ r ≥ 0.31; little or none r ≥ 0.3 (9). P = 0.05 was accepted as the level of significance.

Results

ADC and CV measurements of the internal references

The ADC, ICCs between observers, ROI size, and CV of the internal references are summarized in Table 3. The highest mean ADC value was observed in the gallbladder (3.11 × 10⁻³mm²/s) and the lowest mean ADC value was observed in the spleen (1.04 × 10⁻³mm²/s). ICCs of the ADC measurement indicated an excellent relation between the observers (≥0.91) for all internal references including normal parenchyma (Table 3). The highest CV was detected for spinal cord (17%) and the lowest CV was found for gallbladder bile (8.9%). The overall CV for all internal references, including normal parenchyma, was 14.1%.

Table 3.

Measured parameters of the internal references.

Parameter	Spleen	Gallbladder bile	Kidney cortex	Paraspinal muscle	Spinal cord	CSF
ADC	1.78 ± 0.29	1.04 ± 0.13	3.11 ± 0.28	2.09 ± 0.27	1.68 ± 0.27	1.29 ± 0.22	2.99 ± 0.44
ICC*	0.977	0.952	0.985	0.984	0.949	0.925	0.987
Mean CV (median)	0.16 (0.15)	0.13 (0.12)	0.09 (0.08)	0.13 (0.12)	0.16 (0.15)	0.17 (0.15)	0.15 (0.13)
ROI (mm²)	201 (198–203)	200 (198–203)	197 (146–203)	200 (198–203)	200 (179–212)	22 (4–41)	64 (20–103)
n	74	89	79	89	89	89	89

Results are expressed as the mean ± SD or the mean (range) unless otherwise stated.

ADC (×10⁻³mm²/s) was calculated with multiple b values (0, 600, 800, 1000) for diffusion-weighted imaging.

ADC, apparent diffusion coefficient; CSF, cerebrospinal fluid; CV, coefficient of variation (SD/mean), expressed as a percentage; ICC, interclass correlation; LRD, lesion reference distance; ROI, region of interest.

ICC between observers.

Differentiation of malignant and benign lesions

Results of the evaluation of lesion ADC values and normalized ADCs using various internal references for estimating malignity and benignity are summarized in Fig. 3. The mean ADC values for malignant lesions (1.33 × 10⁻³mm²/s) were significantly lower than those for benign lesions (1.97 × 10⁻³mm²/s) (P < 0.001). There was no significant difference between the benign and malignant groups (P = 0.106) in regard to the mean ADC values for normal parenchyma in lesion-detected organs.

Fig. 3.

Lesion ADC values (×10⁻³mm²/s) and normalized ADCs for malignant and benign abdominal lesions are compared.

ROC analysis

The ROC analysis results of the ADC values or normalized ADCs for differentiating between malignant and benign abdominal lesions are summarized in Fig. 4 and Table 4. The ADC and normalized ADC cutoff values and the related sensitivities, specificities, and accuracies are summarized in Table 4. The normalized ADCs that were obtained using normal parenchyma were better than lesions’ ADCs for differentiating between malignant and benign abdominal lesions Using lesions’ ADCs with a selected set of b values, we found that a cutoff ADC value of 1.455 × 10^–3mm²/s (area under the curve [AUC] = 0.756) enabled this discrimination with 78% sensitivity, 72% specificity, and 76% accuracy. Using normalized ADCs obtained by the internal reference of normal parenchyma showed higher AUC (0.808), a cutoff value of 0.946 permitted this distinction with 78% sensitivity, 62% specificity, and 75% accuracy (or a cutoff value of 1.046; with 67% sensitivity, 95% specificity, and 72% accuracy). Normalized ADCs using CSF showed a similar AUC (0.743) and higher accuracy than that of lesions’ ADC s (Table 4). The normalized ADCs that were obtained using spinal cord and kidney cortex showed similar AUCs and accuracies to those of lesions’ ADCs in differentiating malignant abdominal lesions from benign ones (Table 4). There were significant differences between the normalized ADC obtained using the normal parenchyma and gallbladder bile (P < 0.0001), normal parenchyma and kidney cortex (P < 0.001), and normal parenchyma and spleen (P < 0.001). Significant difference was not found between normalized ADCs of normal parenchyma and the other ones obtained using CSF, spinal cord, and paraspinal muscle.

Fig. 4.

ROC analysis of a lesion’s ADC values and normalized ADCs for estimating malignity/benignity of abdominal lesions.

Table 4.

ROC analysis of ADCs and normalized ADCs of abdominal lesions for estimating benignity/malignity.

Lesion ADC* or normalized ADCs	AUC	SE	P	Asymptotic 95% CI		Cutoff	Sensitivity (%)	Specificity (%)	Accuracy (%)
Lesion ADC* or normalized ADCs	AUC	SE	P	Lower bound	Upper bound	Cutoff	Sensitivity (%)	Specificity (%)	Accuracy (%)
Lesion ADC	0.756	0.059	<0.001	0.640	0.872	1.455	78	72	76
Lesion/normal parenchyma	0.808	0.065	<0.0001	0.681	0.936	0.943 1.046	78 67	62 95	75 72
Lesion/spleen	0.600	0.076	0.188	0.451	0.749	1.214	72	41	66
Lesion/CSF	0.743	0.058	<0.001	0.628	0.857	0.496	83	71	80
Lesion/gallbladder	0.723	0.060	0.003	0.605	0.841	0.492	72	61	70
Lesion/spinal cord	0.716	0.067	0.005	0.584	0.847	1.191	72	76	73
Lesion/paraspinal muscle	0.641	0.061	0.067	0.522	0.760	0.915	72	59	70
Lesion/kidney cortex	0.722	0.060	0.003	0.605	0.839	0.704	72	67	71

P values indicate statistical significance between ADCs or normalized ADCs of benign and malignant lesions.

× 10⁻³mm²/s.

ADC, apparent diffusion coefficient (generated using b values of 0, 600, 800, and 1000 s/mm²); AUC, area under the curve; ROC, receiver operator characteristics; SE, standard error.

Discussion

In this study, the use of normal parenchyma as an internal reference for comparison standards in DWI evaluations increased the diagnostic accuracy when differentiating malignant and benign abdominal lesions. The normalized ADCs obtained using normal parenchyma were better than the lesion ADCs for distinguishing between malignant and benign abdominal lesions. The normalized ADCs obtained using CSF had higher accuracy than lesion ADCs for differentiating between malignant and benign abdominal lesions.

In DWI, the normal parenchyma, especially surrounding the lesion such as liver, kidney and pancreas, can be more useful as an internal reference because of providing visual and numerical comparability. The parenchyma of a lesion-detected organ may not be normal, such as a cirrhotic liver or a liver with steatosis. Most of these changes can be detected using routine MRI sequences. Some changes may be subtle, such as liver fibrosis, and could influence DWI and ADC values. Use of the patient’s clinical information could be helpful for avoiding this situation. The conditions of the patients, such as hyperhydration or dehydration, can affect the measurement. Although ADC measurement can be difficult because of the small sectional area of the spinal cord and CSF, they are the most suitable internal references for use because of their stable positions. The perfusion factor is included in the ADC of abdominal organs and this affects the normalization of ADC. Fluids such as CSF do not need to consider the perfusion effect. In this manner, we believe that CSF is optimal for normalization. CSF can be used as a generalizable standard varying in plane reference structure.

The solid or tissular part of the lesion should be measured because the solid components have typical diffusion properties. Pure cystic lesions were excluded in this study because they can also be readily differentiated from malignant lesions using routine MRI sequences. Abscesses have low ADC values in their central, non-enhancing necrotic viscous contents (10). Patients with a liver abscess, pyelonephritis, diffuse interstitial nephritis, and/or diffuse pancreatitis were excluded from this study because these lesions can be readily differentiated from malignant lesions based on clinical findings and typical appearances on routine MRI sequences (8,10,11).

Several studies, however, have demonstrated important differences in the results when ADCs are measured on different MRI scanners and among results obtained by independent measurements for the same patients (5,12,13). Although artificial phantoms can be used for calibration and standard ADC measurements (14 –18), simulation of tissues is difficult. ADC phantoms commonly have only one or a few diffusion coefficients, have no perfusion component, and may be affected by room temperature (7,15,16). Braithwaite et al. measured the ADC of five intra-abdominal anatomical locations at various times in healthy volunteers (5). Although they found a significant difference between the ADC values of different anatomical locations, they demonstrated no significant difference between ADC values at different times at the same anatomical locations (5). They reported a mean CV of 14%, with no significant differences in CVs between different times or anatomical locations. Corona-Villalobos et al. (7) reported an 11% mean CV for soft tissues and recommended that paraspinal muscle be used as an internal reference. Although paraspinal muscle ADCs can be measured readily on the same slice of the lesion and are not affected by signal differences between slices, its normalized ADC has lower accuracy. In addition, we found that the paraspinal muscles’ CV was higher (16%) than their result. This difference in the results between our study and that of Corona-Villalobos et al. (7) may be related to the imaging parameters or the atrophic changes and fatty content of paraspinal muscles in our study group, which resulted in greater inhomogeneity in our patient population. Paraspinal muscles may be used as a reference standard if atrophic changes are not prominent. Leitão et al. (19) showed that the presence of fat droplets in the liver and lipid emulsion-based phantoms cause a decrease in apparent and pure diffusion coefficients. Repeatability of reference organs and body contents or phantoms is good and is not affected by imaging protocols (5,7). Repeatability is much better than reproducibility (20). The procedures should be adapted to standardized measurements obtained by different researchers and clinicians (21). Reported CVs were higher for liver and spleen (6,17). We found that the reproducibility of measured ADCs was excellent for all internal references. This result may be related to standardized measurements among observers. Although spleen can be easily used as an internal reference with acceptable reproducibility, our ROC analysis showed that its accuracy is lower than that for lesion ADCs. In our study, the CV of gallbladder bile is lowest and the related normalized ADC is similar to the lesion ADC. The measurement technique requires that sediment in the gallbladder should be avoided.

Measured ADC values are changed by applying selected b values, by the echo time used, or they depend on the field strength of the MRI unit (22,23). Although we used a single set of b values, to increase the reproducibility of measured ADCs, we used 600 s/mm² and higher multiple b values as recommended (21,23). A functional diffusion map (fDM) may be used for more accurate evaluation of tumor response and treatment-induced diffusion changes (24 –26).

This study has several limitations. It would be helpful to test the repeatability of internal reference standards by using different protocols or MRI units from different vendors. A recent report showed a good CV for ADC measurements using a temperature-controlled phantom in different MRI scanners (27). Liver DWI and ADC values are especially susceptible to free-breathing and cardiac motion artifacts. Although we could not perform the measurements on different acquisitions or at different sessions, previous reports showed no significant difference between repeated acquisitions or repeated measurements during the same session or different sessions (5,7). Although different kinds of pathology were included, our range of pathology was similar with daily practice of abdominal radiology.

In conclusion, certain body contents or tissues can be used as internal references for comparison. The normal parenchyma, if present, from a lesion-detected organ can be used as an internal comparison standard for DWI. Related normalized ADC is better than lesion ADC for differentiating malignant and benign abdominal lesions. In particular, the normal parenchyma of the liver, kidney, and pancreas surrounding the lesion would be recommendable for comparison standard. CSF can be used as a generalizable standard or varying in plane reference structure.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Yoshikawa

Kawamitsu

Mitchell

. ADC measurement of abdominal organs and lesions using parallel imaging technique. Am J Roentgenol 2006; 187: 1521–1530.

Schmid-Tannwald

Oto

Reiser

. Diffusion weighted MRI of the abdomen: current value in clinical routine. J Magn Reson Imaging 2013; 37: 35–47.

Padhani

Liu

Koh

. Diffusion-weighted magnetic resonance imaging as a cancer biomarker: consensus and recommendations. Neoplasia 2009; 11: 102–125.

Eisenhauer

Therasse

Bogaerts

. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). 2009; 45: 228–247.

Braithwaite

Dale

Boll

. Short- and midterm reproducibility of apparent diffusion coefficient measurements at 3.0-T diffusion-weighted imaging of the abdomen. Radiology 2009; 250: 459–465.

Donati

Chong

Nanz

. Diffusion-weighted MR imaging of upper abdominal organs: field strength and intervendor variability of apparent diffusion coefficients. Radiology 2014; 270: 454–463.

Corona-Villalobos

Pan

Halappa

. Agreement and reproducibility of apparent diffusion coefficient measurements of dual-b-value and multi-b-value diffusion-weighted magnetic resonance imaging at 1.5 Tesla in phantom and in soft tissues of the abdomen. J Comput Assist Tomogr 2013; 37: 46–51.

Koc

Erbay

. Optimal b value in diffusion-weighted imaging for differentiation of abdominal lesions. J Magn Reson Imaging 2014; 40: 559–66.

Choi

Peters

Mueller

. Correlation analysis of ordinal data: from Pearson’s r to Bayesian polychoric correlation. Asia Pacific Educ Rev 2010; 11: 459–466.

10.

Battal

Kocaoglu

Akgun

. Diffusion-weighted imaging in the characterization of focal liver lesions: efficacy of visual assessment. J Comput Assist Tomogr 2011; 35: 326–331.

11.

Wang

Chen

Nikolaidis

. Diffusion-weighted magnetic resonance imaging of pancreatic adenocarcinomas: association with histopathology and tumor grade. J Magn Reson Imaging 2011; 33: 136–142.

12.

Sasaki

Yamada

Watanabe

. Variability in absolute apparent diffusion coefficient values across different platforms may be substantial: a multivendor, multi-institutional comparison study. Radiology 2008; 249: 624–630.

13.

Kivrak

Paksoy

Erol

. Comparison of apparent diffusion coefficient values among different MRI platforms: a multicenter phantom study. Diagn Interv Radiol 2013; 19: 433–437.

14.

Gatidis

Schmidt

Martirosian

. Development of an MRI phantom for diffusion-weighted imaging with independent adjustment of apparent diffusion coefficient values and T2 relaxation times. Magn Reson Med 2014; 72: 459–463.

15.

Lavdas

Behan

Papadaki

. A phantom for diffusion-weighted MRI (DW-MRI). J Magn Reson Imaging 2013; 38: 173–179.

16.

Malyarenko

Galbán

Londy

. Multi-system repeatability and reproducibility of apparent diffusion coefficient measurement using an ice-water phantom. J Magn Reson Imaging 2013; 37: 1238–1246.

17.

Miquel

Scott

Macdougall

. In vitro and in vivo repeatability of abdominal diffusion-weighted MRI. Br J Radiol 2012; 85: 1507–1512.

18.

Chenevert

Galbán

Ivancevic

. Diffusion coefficient measurement using a temperature-controlled fluid for quality control in multicenter studies. J Magn Reson Imaging 2011; 34: 983–987.

19.

Leitão

Doblas

d’Assignies

. Fat deposition decreases diffusion parameters at MRI: a study in phantoms and patients with liver steatosis. Eur Radiol 2013; 23: 461–467.

20.

Colagrande

Pasquinelli

Mazzoni

. MR-diffusion weighted imaging of healthy liver parenchyma: repeatability and reproducibility of apparent diffusion coefficient measurement. J Magn Reson Imaging 2010; 31: 912–920.

21.

Kim

Lee

Park

. Reproducibility of measurement of apparent diffusion coefficients of malignant hepatic tumors: effect of DWI techniques and calculation methods. J Magn Reson Imaging 2012; 36: 1131–1138.

22.

Dale

Braithwaite

Boll

. Field strength and diffusion encoding technique affect the apparent diffusion coefficient measurements in diffusion-weighted imaging of the abdomen. Invest Radiol 2010; 45: 104–108.

23.

Schmidt

Gatidis

Schwenzer

. Impact of measurement parameters on apparent diffusion coefficient quantification in diffusion-weighted-magnetic resonance imaging. Invest Radiol 2015; 50: 46–56.

24.

Ellingson

Malkin

Rand

. Validation of functional diffusion maps (fDMs) as a biomarker for human glioma cellularity. J Magn Reson Imaging 2010; 31: 538–548.

25.

Lee

Bradley

Hussain

. A feasibility study evaluating the functional diffusion map as a predictive imaging biomarker for detection of treatment response in a patient with metastatic prostate cancer to the bone. Neoplasia 2007; 9: 1003–1011.

26.

Reischauer

Gutzeit

Vorburger

. Optimizing the functional diffusion map using Monte Carlo simulations. J Magn Reson Imaging 2012; 36: 1002–1009.

27.

Winfield

Collins

Priest

. A framework for optimization of diffusion-weighted MRI protocols for large field-of-view abdominal-pelvic imaging in multicenter studies. Med Phys 2016; 43: 95–95.