Apparent Diffusion Coefficient (ADC) Measurements in Pancreatic Adenocarcinoma: A Preliminary Study of the Effect of Region of Interest on ADC Values and Interobserver Variability Chao Ma, MS,1 Li Liu, MD,1 Jing Li, MD,1 Li Wang, MD,1 Lu-guang Chen, MS,1 Yong Zhang, PHD,2 Shi-yue Chen, MS,1 and Jian-ping Lu, MD1* Purpose: To assess the influence of region of interest (ROI) on tumor apparent diffusion coefficient (ADC) measurements and interobserver variability in pancreatic ductal adenocarcinoma (PDAC). Materials and Methods: Twenty-two patients recruited with pathology-proven PDAC underwent diffusion-weighted imaging (DWI, 3.0T) prior to the surgical resection. Two independent readers measured tumor ADCs according to three ROI methods: whole-volume, single-slice, and small solid sample of tumor. Minimum and mean ADCs were obtained. The interobserver variability for each of the three methods was analyzed using interclass correlation coefficient (ICC) and Bland–Altman analysis. The minimum and mean ADCs among the ROI methods were compared using nonparametric tests. Results: The single-slice ROI method showed the best reproducibility in the minimum ADC measurements (mean difference 6 limits of agreement between two readers were 0.025 6 0.25 3 1023 mm2/s; ICC, 0.92) among the three ROI methods. For the solid tumor sample ROI, both minimum ADC and mean ADC measurements reproducibility were the worst, with limits of agreement up to 60.50 3 1023 mm2/s and 60.32 3 1023 mm2/s, respectively (ICCs, 0.41/0.58). Both the minimum and mean ADCs demonstrated significant differences among the three ROI methods (both P < 0.001). The post-hoc analyses results showed no significant difference with regard to the mean ADCs between whole-volume and single-slice ROI methods (P 5 0.14). Conclusion: The ROI method had a considerable influence on both the minimum and mean ADC values and the interobserver variability in PDAC. The worst interobserver variability was observed for both the minimum and mean ADCs derived from small solid-sample ROI. J. MAGN. RESON. IMAGING 2016;43:407–413.
agnetic resonance imaging (MRI) is an important tool to detect and differentiate pancreatic diseases. Specifically, diffusion-weighted imaging (DWI) provides an additional promising dimension to the conventional anatomical MRI.1 Notably, several recent studies have indicated the underlying value of DWI in the studies of pancreatic diseases; for instance, significantly lower apparent diffusion
coefficient (ADC) was observed in pancreatic cancer than that in benign pancreas tissue.2–15 ADC values reflect the physical properties of the tissues of interest, which can be influenced by the placements of the region of interest (ROI). In tumor studies, three major ROI methods have been applied to achieve ADC values, including whole volume,17,18 single slice,8,10 and small
View this article online at wileyonlinelibrary.com. DOI: 10.1002/jmri.25007 Received Jan 19, 2015, and in revised form Jun 30, 2015. Accepted for publication Jun 30, 2015. The first two authors contributed equally to this work. *Address reprint requests to: J-P.L., Department of Radiology, Changhai Hospital of Shanghai, Second Military Medical University, No. 168 Changhai Road, Shanghai 200433, China. E-mail: [email protected]
From the 1Department of Radiology, Changhai Hospital of Shanghai, Second Military Medical University, Shanghai, China; and 2GE Healthcare, MR Group, Shanghai, China
C 2015 Wiley Periodicals, Inc. V 407
Journal of Magnetic Resonance Imaging
TABLE 1. Main Parameters for All Applied MR Sequences
Thickness/ gap (mm)
Flip angle (0)
2D Single-Shot Fast Spin Echo, SSFSE (MRCP)
Axial Fast Spin Echo, FSE (T2WI)
Axial Single-Shot Echo Planar Imaging, ss DWEPI (DWI)
3D fat-suppressed Gradient Echo, 3D GRE (LAVA)
NEX 5 2 for DWI at b0, NEX 5 4 for DWI at b600.
sample ROI of tumor.11–18 The assessment of ADC reproducibility is actually an important point of research in body MRI. Some studies have been carried out to evaluate the influence of ROI sampling on liver ADC 19 and ADC reproducibility in spleen.20 However, it is debatable if ROIs could ideally incorporate the entire tumor volume or only a representative tumor section of pancreatic cancer. We chose to study the impact of the ROI method selection on pancreatic tumor ADC measurements and on interobserver variability, which play significant roles on investigating the application of ADC as a potential marker to differentiate pancreatic cancer from normal pancreas.
Materials and Methods Patients This prospective study was approved by our Institutional Review Board. Signed written informed consent was obtained from all participants after detailed explanation but prior to imaging. Fifty-one patients with known (surgically pathological proven) pancreatic ductal adenocarcinomas (PDAC) were enrolled in the study between February 2014 and October 2014. Twenty-nine PDAC patients with unclearly demarcated hyperintensity on DWI images as compared to the surrounding pancreas (according to the definition proposed by Fukukura et al 13) were excluded from the study. The DWI data of the remaining 22 PDAC patients (13 males, 9 females; mean age, 63.7 66.8 years; range 50–76 years) were investigated in the study.
MRI All examinations were performed on a 3.0T MR (Signa HDxt V16.0, GE Healthcare, Milwaukee, WI) with an eight-element phased array coil. All the patients underwent routine conventional MRI protocols and transverse respiratory triggered single-shot echo-planar DWI (weighted along three orthogonal gradient directions), using b values of 0 and 600 s/mm2. Selective presaturation 408
with inversion recovery (SPIR) was used to achieve fat saturation. The main scan parameters and the scanning order of sequences are presented in Table 1. Contrast-enhanced liver acceleration volume acquisition (LAVA) was performed with gadopentetate dimeglumine injection (physiological saline, 10–15 ml; media, 0.2–0.3 ml/ kg) at the end of the study.
Data Analysis Entry and exclusion of patients for analysis was performed by one radiologist who had more than 10 years of experience in abdominal radiology. The reader was informed that the patients had been diagnosed with PDAC but was blinded to the histopathological data at the time of evaluation. DWI images of b600 were analyzed according to the signal intensity 13 of the tumor in comparison with the adjacent normal pancreas tissues. The DWI images with hyperintensity were sufficient to accurately delineate the tumor and to provide the optimal tumor size evaluation, compared to the conventional anatomical MRI sequences.21 Thus, to investigate the validity of tumor ADC, especially for whole volume ROI and the single-slice ROI method, only the DWI images demonstrating hyperintensity with clear borders to the surrounding pancreatic parenchyma were selected and utilized for further analyses. DWI data were processed using a standard software package (Functool 9.4.05, GE AW 4.4, GE Healthcare). ADC values were calculated for all slices voxel-by-voxel based on the achieved DWI images. DWI images of both b0 and b600 and ADC maps were loaded in the software in conjunction. ROIs were drawn on multiple slices of the DWI images of b600 and were directly colocalized on the ADC map. Two radiologists (L.L. and J.L., with a 10-year and 7-year experience in abdominal radiology, respectively) independently measured the tumor ADCs according to all the three ROI methods. Applying the whole-volume method, freehand ROIs were drawn along the border of the high signal of the tumor on DWI images (b600) to cover the entire tumor area on each tumorcontaining slice. The mean ADC value of the whole tumor was regarded as the mean of all voxel ADCs within each recorded ROI Volume 43, No. 2
Ma et al.: Effects of ROI on ADC Measurements in PDAC
FIGURE 1: Representative images obtained from a 63-year-old male patient with adenocarcinoma at the head of the pancreas: clearly demarcated hyperintensity while compared with the surrounding pancreas tissues on DWI images. Axial contrast-enhanced CT image shows a hypoattenuating lesion at the head of the pancreas (A); Axial T2-weighted image (B); contrast-enhanced T1weighted image (C) demonstrates hypovascularity of the mass. For whole-volume and single-slice methods, freehand ROIs were drawn along the high signal intensity border of the tumor on six tumor-containing slices of the obtained DWI images (b600, the middle row of D) to cover the entire tumor. For the solid sample ROI method, tumor ADC was measured by drawing a round- or oval-shaped ROI within the solid tumor areas.
of the entire lesion. Mean ADC of the single-slice method was obtained from the largest observed tumor area slice throughout the whole-volume measurements. For the solid tumor samples, mean ADC was calculated from a largest possible round or oval ROI, which was placed on the solid portion of the tumor (Fig. 1), avoiding pancreatic ducts and cystic lesions by referring to other MRI images such as T2-weighted imaging (T2WI) or LAVA. Minimum ADCs were also obtained for the ROI methods in each case.
Statistical Analysis Statistical analyses were performed by using SPSS software (v. 16.0, SPSS, Chicago, IL). Interobserver variability for tumor ADC measurements for each ROI method was analyzed by interclass correlation coefficient (ICC: 0–0.20, poor correlation; 0.21–0.40, fair correlation; 0.41–0.60, moderate correlation; 0.61–0.80, good correlation; and 0.81–1.00, excellent correlation) 22 and the method of Bland–Altman.23 The minimum and mean ADCs were averaged between the results of two readers for further analyses. Comparisons of the achieved minimum and mean ADCs between three ROI methods were performed by using Friedman tests and posthoc analyses, which were conducted with Wilcoxon signed-rank February 2016
tests and a Bonferroni correction applied.24 The statistical significance threshold of the Friedman test was set at a P below 0.05, while at a P below 0.017 (0.05/3) for post-hoc tests regarding the Bonferroni correction for the three comparisons.
Results Interobserver Variability of ADC Values The typical axial DWI images and ADC maps for ADC measurements with the three ROI methods are demonstrated in Fig. 1. The minimum and mean ADC values of PDAC measured by two readers using three ROI methods are summarized and shown in Table 2. For the minimum ADC values, the mean difference (bias) and the 95% confidence interval of the mean difference (limits of agreement) between two readers were 0.119 [–0.283-0.521] 3 1023 mm2/s for whole-volume ROI (ICC, 0.75), 0.025 [–0.225-0.275] 3 1023 mm2/s for single-slice ROI (ICC, 0.92), and 20.107 [–0.610-0.396] 3 1023 mm2/s for small solid sample ROI (ICC, 0.41). 409
Journal of Magnetic Resonance Imaging
TABLE 2. Interobserver Comparisons
Solid sample ROI
Reader 1 0.87 (0.61, 0.99) [0.03, 1.13]
Reader 2 0.70 (0.45, 0.88) [0.03, 1.08]
Reader 1 0.96 (0.64, 1.13) [0.03, 1.33]
Reader 2 0.89 (0.75, 0.96) [0.03, 1.33]
Reader 1 1.11 (0.96, 1.18) [0.06, 1.38]
Reader 2 1.15 (1.11, 1.25) [0.70, 1.38]
1.46 (1.40, 1.54) [1.19, 1.60]
1.42 (1.34, 1.50) [1.24, 1.68]
1.42 (1.35, 1.52) [1.03, 1.62]
1.39 (1.29, 1.47) [1.20, 1.86]
1.31 (1.21, 1.37) [0.65, 1.52]
1.33 (1.23, 1.39) [0.89, 1.53]
Apparent diffusion coefficient (ADC) measurements (31023 mm2/s) of pancreatic adenocarcinoma obtained by whole volume, a single slice and small sample ROI methods by two independent readers, respectively. a Data are expressed as medians; numbers in parentheses are first quartiles (q1) and third quartiles (q3); numbers in brackets are ranges.
The observed minimum ADCs were the least scattered when achieved with the single-slice method. Graphic illustrations of these data with Bland–Altman plots are displayed in Fig. 2. For the mean ADC values, the mean difference (bias) and limits of agreement between two readers were 0.022 [– 0.135-0.180] 3 1023 mm2/s for whole-volume ROI (ICC, 0.78), 0.004 [–0.267-0.274] 3 1023 mm2/s for single-slice ROI (ICCs, 0.59), and 0.023 [–0.343-0.297] 3 1023 mm2/s for small solid sample ROI (ICC, 0.58). The mean ADCs in small solid sample ROI were more scattered than the other two ROI methods. Graphic illustrations of these data with Bland–Altman plots are displayed in Fig. 3. Comparison of ADC Values Between Three ROI Methods Comparisons of the minimum and mean ADCs of PDAC with whole-volume, single slice and solid sample ROIs from the 22 patients are summarized and presented in Table 3. Friedman test results demonstrated significant differences among the minimum and mean ADCs of the three methods (both P < 0.001). Post-hoc analyses with Wilcoxon signed-
rank test results indicated a statistically significant lower mean ADCs in solid tumor sample than that in wholevolume or single-slice (both P < 0.001). However, no significant difference was observed for the mean ADCs between whole-volume and single-slice methods (P 5 0.14). In addition, a significant difference was observed for the minimum ADCs in the comparison of any two methods (P < 0.001).
Discussion Our results demonstrate that the minimum and mean ADCs of tumor and interobserver variability are dependent on the ROI methods in PDAC patients. The reproducibility of mean ADC of PDAC in the whole volume and single slice of tumor are acceptable, considering the fact that the mean interobserver bias of ADC measurements did not exceed 60.10 3 1023 mm2/s and the limits of agreement were less than 60.30 3 1023 mm2/s. The mean or minimum ADC measurements were used to investigate pancreatic adenocarcinoma; however, few studies have demonstrated the reliability of the ADC results derived from any of the three ROI methods in PDAC. A previous study assessed the influence of the three ROI methods on
FIGURE 2: Interobserver reproducibility of the minimum ADC measurements (31023 mm2/s) of all three ROI methods in PDAC. For Bland–Altman plots: the difference of minimum ADC measurements (y-axis) were plotted against the mean minimum ADCs (xaxis), with mean absolute difference (bias) (continuous line) and 95% confidence interval of the mean difference (limits of agreement) (dashed lines).
Volume 43, No. 2
Ma et al.: Effects of ROI on ADC Measurements in PDAC
FIGURE 3: Interobserver reproducibility of the mean ADC measurements (31023 mm2/s) of all three ROI methods in PDAC. For the Bland–Altman test, the difference of mean ADC measurements (y-axis) were plotted against the mean ADCs (x-axis), with mean absolute difference (bias) (continuous line) and 95% confidence interval of the mean difference (limits of agreement) (dashed lines).
tumor ADC measurements and the interobserver variability in patients with local advanced rectal cancer.25 It was demonstrated that the ROI size and positioning showed considerable influence on the tumor mean ADC values and interobserver variability as well. They concluded that wholevolume measurements provided the most reproducible results. Another study investigated the influence of different shaped ROIs on tumor ADC measurement and the corresponding interobserver variability in endometrial carcinoma,26 which indicated that ROI shape had no significant influence on the tumor mean ADC values for the endometrial carcinoma existing in the intrauterine cavity when the tumor shape was approximate to oval or round. In addition, to the best of our knowledge, our study is the first to investigate the ADC measurement using whole-volume ROI in PDAC, as well as comparing the influence of different ROI methods on tumor ADC measurements. It was notable that good interobserver correlation of ADC measurements did not indicate that the two measurements were consistent.23 This was confirmed by the unacceptable minimum ADC measurements reproducibility with limits of agreement up to 60.40 3 1023 mm2/s, whereas the measurements showed good correlation (ICC, 0.75) at the same time. The minimum ADCs differed significantly between any two methods. Furthermore, the mean minimum ADC values were lowest for whole-volume ROI freehand ROIs
drawn along the border of the tumor on DWI images to cover the entire tumor area on each tumor-containing slice in the whole-volume method. Thus, it would contain the areas with highest cellular density in the tumor, which may not be covered by either single-slice or solid-sample ROI methods. The small solid-sample ROI method resulted in significantly lower mean ADC values than both whole-volume and single-slice ROIs. The small-sample ROIs only included the most viable solid tumor parts, which may explain the observed lowest mean ADC values. In addition, the vessels, ducts, and necrosis were likely to be excluded from the solid-sample ROIs, whereas these factors would influence tumors ADC values in vivo. Thus, the relatively higher ADCs obtained by whole-volume or single-slice ROI methods could reflect the heterogeneous nature of the tumors, which included solid lesion, ducts, and fibrosis. In clinical use, the solid-sample ROI method was frequently used to achieve ADC in PDAC, and some previously studies have reported that the significantly lower ADC was observed in PDAC than that in benign pancreas tissue.2–7,12–16 In the current study, two readers independently measured the small solid-sample tumor ADCs, without consensus selection of the same slice for each case. Despite significant lower mean ADCs of solid tumor sample in PDAC, the least interobserver variability was also observed in this method for the
TABLE 3. Comparisons of Minimum and Mean ADCs (6 Standard Deviation, SD) Measured From the Whole Volume, a Single Slice and Small Sample of the Pancreatic Adenocarcinoma.
Parameters Minimum ADC (31023 mm2/s) 23
Mean ADC (310
Total ROI size (mm )
Solid sample ROI
0.70 6 0.29
0.84 6 0.31a
1.09 6 0.18a,b