To evaluate the utility of diffusion-weighted magnetic resonance imaging (DWI) in pancreatic ductal adenocarcinoma with various grades of differentiation.
To evaluate the utility of diffusion-weighted magnetic resonance imaging (DWI) in pancreatic ductal adenocarcinoma with various grades of differentiation.
Following Institutional Review Board (IRB) approval, 21 consecutive patients with surgical pathology-proven pancreatic adenocarcinomas were retrospectively evaluated. Histopathologic characteristics and grades of differentiation of adenocarcinomas were analyzed. Twenty-one patients without a known history of pancreatic disease were evaluated as the control group. Anatomic MR images and DW images were acquired using 1.5-T MR systems. DWI with b values of 0 and 500 sec/mm2 were performed on both patients and control groups. The difference in mean apparent diffusion coefficient (ADC) values among groups of normal pancreatic parenchyma, adenocarcinomas with poor differentiation, and adenocarcinomas with well/moderate differentiation were compared using one-way analysis of variance.
Mean ADCs of pancreatic adenocarcinomas (1.77 ± 0.45 × 10−3mm2/sec) was not significantly lower than that of normal parenchyma (1.98 ± 0.31) (P = 0.09). When adenocarcinomas were subdivided based on grades of differentiation, however, poorly differentiated adenocarcinoma with histopathologic characteristics of limited glandular formation and dense fibrosis had significantly lower ADCs (1.46 ± 0.17) compared to those of well/moderately differentiated adenocarcinomas (2.10 ± 0.42) characterized by neoplastic tubular structures (P < 0.01). Well/moderately differentiated adenocarcinomas with dense fibrosis showed significantly lower ADC values (1.49 ± 0.19) than those with loose fibrosis (2.26 ± 0.30) (P = 0.01).
Difference in ADC values using DWI between poorly and well/moderately differentiated pancreatic ductal adenocarcinoma may relate to differences in glandular formation and density of fibrosis. J. Magn. Reson. Imaging 2011;33:136–142. © 2010 Wiley-Liss, Inc.
PANCREATIC DUCTAL ADENOCARCINOMA is the fourth most common cause of cancer-related death (1, 2). Prognosis and survival rates in patients with pancreatic adenocarcinoma greatly depend on tumor size, resection margin status, lymph node status, and grade of differentiation (3, 4). Early detection and identification of the aggressiveness of pancreatic adenocarcinoma would be helpful in the development of more effective therapeutic strategies (5, 6). Diffusion-weighted imaging (DWI) has been increasingly used as an imaging method to evaluate various cerebral and extracerebral tumors, and more recently to predict their grade of differentiation (7–9). In theory, DWI provides qualitative and quantitative information pertinent to biological tissues which is based on measurement of thermally induced diffusivity of water molecules shown by apparent diffusion coefficient (ADC) (10, 11). Low ADC values can be attributed to a greater cellular structure or extracellular fibrosis (9–11). Thus, certain tumors, such as gliomas, with poor differentiation have been distinguished from well-differentiated tumors by decreased ADC values on DWI as a result of dense cellularity (7, 8, 12).
Recently, DWI has been used as a complementary imaging technique to conventional magnetic resonance imaging (MRI) to diagnose pancreatic adenocarcinoma. High sensitivity (96.2%) and specificity (98.6%) in the detection of pancreatic adenocarcinoma have been reported utilizing DWI (13). However, conflicting results for ADC values have been described with the use of DWI for characterization of adenocarcinoma. Both lower and higher ADC values have been reported in pancreatic adenocarcinomas compared to the surrounding normal pancreatic parenchyma (14–16). These reports, however, have not accounted for the effect of tumor grade of differentiation and histopathological correlation for the ADC values.
The purpose of our study was: 1) to evaluate the utility of ADC values in distinguishing pancreatic ductal adenocarcinoma with various grades of differentiation; and 2) to identify histopathological characteristics associated with the variable ADC values of pancreatic adenocarcinoma.
This study was approved by our Institutional Review Board (IRB) and was compliant with the Health Insurance Portability and Accountability Act (HIPAA). A waiver of informed consent was obtained for this retrospective study. We reviewed our MRI database from June 2006 to July 2008 and identified 33 patients with surgically proven and histopathologically analyzed pancreatic adenocarcinoma who underwent MRI. Our inclusion criteria were: 1) MR examinations including DWI obtained before surgery; 2) pancreatic masses larger than 1 cm at anatomical MR to allow accurate ADC measurements; and 3) MR and DW images of adequate quality without significant artifacts. Twelve patients were excluded due to the following reasons: lesions were too small (<1 cm) in five patients; image quality was unsatisfactory, caused by respiratory and motion artifacts in four patients; and a complete DWI exam was not performed in three patients. The final cohort consisted of 21 consecutive patients (12 male and 9 female; mean age 68 years; range 48–86 years) with pathologically proven pancreatic ductal adenocarcinoma.
Our control group for the evaluation of normal ADC values in the pancreas consisted of 21 patients with suspected renal lesions referred to our radiology department for MR. These patients (12 male and 9 female; mean 57 years; range 34–87 years) did not have any documented history of pancreatic disease; all had normal serum amylase and lipase levels.
MR examinations were performed using 1.5-T MR systems (Magnetom Espree Siemens Healthcare, Erlangen, Germany). All patients were examined in the supine position. Six-element phased array matrix coil (anterior) and six elements of spine matrix coil (posterior) were used for signal reception.
The anatomical abdominal MRI protocol for imaging the pancreas used in our institution included transverse and coronal T2-weighted half-Fourier acquisition single-shot turbo spin echo (HASTE), and T2-weighted turbo spin-echo (TSE) with fat suppression. MR cholangiopancreatography sequences (MRCP) were performed with axial and coronal breathold 2D half-Fourier rapid acquisition with relaxation enhancement (RARE) and 3D navigator triggered MRCP. Unenhanced and dynamic gadolinium-enhanced T1-weighted images were acquired utilizing fat suppressed gradient echo using shared prepulses (SHARP). T1-weighted fat suppressed gradient echo images were obtained using the following parameters: TR/TE, 120–160/1.9 msec; flip angle, 80°; slice thickness, 6 mm; gap, 1.8 mm; matrix, 256 × 179. Dynamic gadolinium-enhanced MR images were acquired in the arterial phase (scan time based on fluoroscopy-preparation timing sequence), venous phases (45–60 and 90 seconds), and delayed phase with images obtained 2–5 minutes after contrast injection. Gadopentetate dimeglumine (Magnevist; Bayer HealthCare Pharmaceuticals, Berlin, Germany) was administered at a dose of 0.1 mmol/kg, followed by 20 mL saline flush (2 mL/sec) with a power injector (Spectris; Medrad, Warrendale, PA).
DW MRI was obtained by single-shot spin-echo echo-planar imaging (EPI) with spectral presaturation attenuated inversion-recovery (SPAIR) fat-suppressed pulse. Integrated parallel imaging techniques (iPAT) using generalized autocalibrating partially parallel acquisitions (GRAPPA) were used with a 2-fold acceleration. The parameters were: TR/TE, 5000/80 msec; matrix of 156 × 192; bandwidth, 1446 Hz/px; section thickness, 6 mm; gap, 1.8 mm; field of view (FOV), 300–400 mm; partial Fourier factor 6/8; averages, 2; parallel imaging factor 2; free breathing; b values, 0 and 500 sec/mm2. Typical scanning time was less than 2 minutes. All separate image series were acquired with diffusion weighting in the axial direction using tridirectional diffusion gradients. The total slices of 2D DW images varied as the liver and pancreas were covered in all patients; ≈30 slices were acquired when performing DW images.
Two radiologists, with 8 and 18 years of experience in body MR, reviewed each study retrospectively in consensus. In order to evaluate intraobserver reproducibility, the radiologists reevaluated all studies in consensus after 3 months. The radiologists were aware of the location of the tumor based on anatomic images and were blinded to the histopathological results.
Pancreatic ductal adenocarcinoma was seen as hypointense mass on precontrast T1-weighted images with fat suppression, and showed decreased enhancement relative to normal pancreatic parenchyma on the arterial phase and progressive enhancement on delayed sequences (17). Cystic lesions within the tumor were identified as areas that appeared hypointense on precontrast T1-weighted and hyperintense on T2-weighted images and did not enhance on the dynamic enhanced MR images; lack of enhancement was confirmed with region of interest (ROI) measurements.
In brief, a pair of DW images was acquired with b values of 0 (b0) and 500 sec/mm2 (b1). ADC values were calculated based on the formula: ADC (× 10−3mm2/sec) =−(1/(b1−b0)) ×ln(S1/S0) × 1000, where S0 and S1 were the signal intensity of tumor measured on DW images with b0 and b1, respectively.
A round- or oval-shaped ROI was placed on the solid and homogeneous portion of the tumor, which was identified on the corresponding anatomic MR images. The size of the ROI was as large as possible and was adjusted according to the size of each tumor. Based on the anatomic MR images, any cystic portions within the tumor were avoided.
For the normal control group, a round- or oval-shaped ROI was drawn by the same reviewer. ROIs covered the largest possible area over the pancreas head, body, and tail. ADC values were obtained from the head, body, and tail of the pancreas and were averaged and then used as representative values for normal parenchyma.
Pathological evaluation of all tumors was made on surgically resected specimens. The median time interval between MRI and surgical resection was 8 days with a range of 3–21 days. Hematoxylin and eosin (H&E)-stained slides were reviewed using a microscope (Olympus BX51; Olympus, Tokyo, Japan) and selected images were taken using a digital camera (Olympus DP71; Olympus). All histopathological analyses were performed independent of the DWI study results by one attending subspecialized pathologist with 8 years of experience specifically for this study to evaluate features that affect DWI. After the MR images and pathologic slides were independently reviewed, concordance of the location of the tumor on MRI and associated histology slides was ensured in consensus by the pathologist and one of the radiologists in this study.
Based on classification system of the World Health Organization (WHO) and practical grading scheme for pathology (18), all cases evaluated in the present study were subcategorized as well, moderately, and poorly differentiated adenocarcinomas. For the purpose of this study critical morphological features were recorded for each case including glandular differentiation, mucin production, grade of nuclear atypia, and mitotic activity (18). Glandular differentiation in pancreatic ductal adenocarcinoma was graded as involving the majority of tumor (>75%), moderate degree (10%–75%), and little or none (<10%) (19). The final grade of each tumor represented the highest grade observed, excluding the area of invasive front where individual cells were often identified.
Because pancreatic ductal adenocarcinoma is frequently related to an intense fibrotic process known as desmoplastic reaction (20), density of fibrosis was semiquantitatively measured in addition to other morphological features. Briefly, representative H&E-stained slides containing tumor were analyzed microscopically. The ratio of tumor and its fibrotic stromal component was examined at 100× magnification level. The density of fibrosis was graded as loose fibrosis (<50% of the tumor volume) and dense fibrosis (≥50% of the tumor volume) in this study (21, 22).
Quantitative variables are reported as mean ± standard deviation (SD) after verifying the absence of substantial departure from the normality of samples data (standardized kurtosis and standardized skewness <2 and nonsignificant Shapiro–Wilks test for normality). Intraobservers variability was evaluated by Bland–Altman plots.
The difference of ADC values between pancreatic adenocarcinoma and normal pancreatic parenchyma was compared by Student's t-test. Because of the rarity of well-differentiated adenocarcinoma and similar clinical outcomes of well and moderately differentiated adenocarcinoma, well (n = 1) and moderately differentiated adenocarcinomas (n = 9) were combined as one group for statistical analysis and comparison (6). ADC values of the three patient groups of normal pancreatic parenchyma, adenocarcinoma with poor differentiation, and adenocarcinoma with well/moderate differentiation were analyzed and compared using one-way analysis of variance (ANOVA). Post-hoc testing was performed using the Tukey-b method.
Estimates and 95% confidence intervals (CI) of P-values were determined using 3000 sample bootstrapped estimation. P < 0.05 was required to reject the null hypothesis. Statistical analysis was performed with MedCalc software for Windows (Mariakerke, Belgium).
Based on the WHO staging criteria, 11 patients with poorly differentiated adenocarcinoma were identified (23, 24). Glandular formation was limited or absent in the poorly differentiated adenocarcinomas (Fig. 1). These tumors characteristically presented as sheets of cell growth or singular and small nests of tumor cells without tubular or ductal formation. An abundant dense fibrotic stroma with a pattern of thick collagen bundles was detected in the 6 of 11 patients with poorly differentiated adenocarcinoma.
One patient with well-differentiated adenocarcinoma and nine patients with moderately differentiated tumors were identified. Both the well and moderately differentiated adenocarcinomas (Fig. 2) showed neoplastic tubular structures with a haphazard distribution, with the well-differentiated adenocarcinoma only showing mild cytologic atypia. Dense fibrosis was detected in one patient with moderately differentiated adenocarcinoma and in one patient with well-differentiated adenocarcinoma.
Mean intraobserver variability for determination of ADC value was 1.25% (95% CI, −3.96% to 6.46%) according to the Bland–Altman plots.
Mean ADC values of pancreatic parenchyma was 1.98 ± 0.31 × 10−3mm2/sec in the control group. Compared to normal pancreatic tissue from the control group, pancreatic adenocarcinoma had lower mean ADC values (1.77 ± 0.45 × 10−3mm2/sec); however, a significant difference was not seen between the two groups (P = 0.09). After subdivision based on grading differentiation, mean ADC values of poorly differentiated pancreatic adenocarcinomas (1.46 ± 0.17 × 10−3mm2/sec) (Fig. 1) was lower than that of well/moderately differentiated adenocarcinoma (2.10 ± 0.42 × 10−3mm2/sec) (Fig. 2) (P < 0.01). Compared to normal parenchyma in the control group, poorly differentiated pancreatic adenocarcinomas had lower mean ADC values (P < 0.01). In contrast, mean ADC values of well/moderately differentiated adenocarcinoma was higher than that of normal tissues from the control group but the difference did not reach statistical significance (P = 0.38). Comparison of the ADC values of normal parenchyma from the control group and pancreatic adenocarcinomas with various grades of differentiation are shown in Fig. 3.
On further analysis, well/moderately differentiated adenocarcinomas with dense fibrosis showed significantly lower ADC values (1.49 ± 0.19 × 10−3mm2/sec) than those with loose fibrosis (2.26 ± 0.30 × 10−3mm2/sec) (P = 0.01). However, a statistically significant difference in ADC values between poorly differentiated adenocarcinoma with dense fibrosis (1.40 ± 0.19 × 10−3mm2/sec) and those with loose fibrosis (1.54 ± 0.14 × 10−3mm2/sec) was not seen (P = 0.21). Distribution of ADC in each subcategory of adenocarcinoma with dense or loose fibrosis is shown as scatter plots in Fig. 4.
As a functional imaging procedure, DWI has been used to detect and characterize tumors in different organs (25), monitor tumor response to treatment (26), and predict grades of malignancy and prognosis (7). In this study a significant difference in ADC values of pancreatic ductal adenocarcinoma with various grades of differentiation was observed and likely accounts for the variable results of the ADC values of pancreatic adenocarcinoma seen in prior studies (14–16).
Glandular formation is the critical morphological characteristic for grading differentiation of ductal adenocarcinoma (19, 27). Poorly differentiated ductal adenocarcinoma characteristically shows limited to no glandular formation. In contrast, neoplastic tubular and duct-like structures with or without mucinous secretions in their lumina are noted in well and moderately differentiated adenocarcinomas. These tubular and duct-like structures contain high fluid content in vivo, which may account for the freedom of diffusion of water molecules and high ADC values (9, 11). This may explain the significant difference in ADC value between the low-grade adenocarcinomas and the high-grade adenocarcinomas observed in this study. Therefore, glandular formation may be one of the histopathological factors that contribute to different ADC values associated with different grades of differentiation of ductal adenocarcinomas.
In addition, pancreatic ductal adenocarcinoma is frequently associated with an intense fibrotic process known as desmoplastic reaction (20). In theory, extracellular fibrosis characterized by extracellular accumulation of collagen, glycosaminoglycans, and proteoglycans might restrict water diffusion, resulting in a relatively low ADC value (9). It has been demonstrated by Muraoka et al (21) that adenocarcinoma with dense fibrosis had lower ADC values (n = 7; 1.01 ± 0.29 × 10−3mm2/sec) than that of tumors with loose fibrosis (n = 3; 1.88 ± 0.39 × 10−3mm2/sec). In agreement with these results, we also found significantly lower ADC values of well/moderately differentiated adenocarcinomas with dense fibrosis compared to those with loose fibrosis. Although the histopathological characteristics of glandular formation should correlate well with more freedom of water diffusion and high ADC values, the appearance of fibrosis within such tumors should account for more restricted diffusivity of water molecules, resulting in low ADC values. Fibrosis may or may not be the predominant feature in ultimately determining the actual ADC value of a tumor. Since poorly differentiated adenocarcinoma characteristically shows limited or no glandular formation, the variability in density of fibrosis may not have as significant an influence on ADC values compared to well and moderately differentiated tumors, as we did not encounter a statistically significant difference in ADC values between poorly differentiated adenocarcinoma with dense or loose fibrosis. In summary, both glandular formation and density of fibrosis can contribute to ADC value of any given tumor, and the delicate interplay of these two factors, along with perhaps more yet-to-be-identified factors, will ultimately determine the overall ADC value of an adenocarcinoma.
Although DW images were acquired using different scanners or protocols, the mean ADC value of normal pancreatic tissues (1.98 ± 0.31 × 10−3mm2/sec) in this study was comparable with reported ADC values of 1.78 ± 0.07 × 10−3mm2/sec by Yoshikawa et al (14), 2.06 ± 0.42 × 10−3mm2/sec by Lee et al (15), 1.76 ± 0.40 × 10−3mm2/sec by Fattahi et al (16), and 1.90 ± 0.06 × 10−3mm2/sec by Matsuki et al (28). However, conflicting ADC values of pancreatic adenocarcinoma have been reported (14–16). Compared to the normal pancreatic parenchyma, some prior reports of pancreatic adenocarcinomas had lower ADC values of 1.46 ± 0.20 × 10−3mm2/sec (15), 1.46 ± 0.18 × 10−3mm2/sec (16), and 1.44 ± 0.20 × 10−3mm2/sec (28), which were similar to ADC values of poorly differentiated pancreatic adenocarcinomas (1.46 ± 0.17 × 10−3mm2/sec) demonstrated by our study. In contrast, Yoshikawa et al (14) reported that pancreatic adenocarcinoma had significantly higher ADC values (2.32 ± 0.26 × 10−3mm2/sec) than normal tissue (1.76 ± 0.40 × 10−3mm2/sec). These results were more consistent with ADC values of well/moderately differentiated adenocarcinoma (2.10 ± 0.42 × 10−3mm2/sec) observed in the present study. A significant difference in ADC values between poorly differentiated adenocarcinomas and their well to moderately differentiated counterparts shown in this study may help explain the contradictory results seen in these previous published studies.
Our study had several limitations. First, the relatively small sample size of our study may have produced a selection bias. Although only one (4%) patient with well-differentiated pancreatic adenocarcinoma was evaluated in our study, this is in keeping with its low reported incidence of 6% (1, 6). As in previous studies, we combined the well and moderately differentiated adenocarcinomas into one group, as the outcome for both groups is similar and significantly better than those with poor differentiation (6). Further studies with larger samples size are needed to confirm our results. Second, similar to previous studies, it is difficult to exactly quantify the degree of fibrosis. Without a well-established and validated grading system, this remains a challenge even when a pathologist carefully reviewed the slides blinded to the imaging results (21, 29). In addition, another limitation involving DWI studies in the abdomen has been the use of different b values for determining ADC values. It is also a limitation that the application of a single monoexponential fit to a decay curve is likely better modeled with biexponential decay.
In conclusion, we demonstrated there was a significant difference in ADC values using DWI between poorly and well/moderately differentiated pancreatic ductal adenocarcinoma, which may relate to the various degrees of glandular formation and density of fibrosis. These differences may account for the conflicting ADC values of pancreatic adenocarcinoma reported in the literature. Further studies of the effects of cellularity, glandular formation, and other histological parameters on ADC values will elucidate the complex histopathological and biophysical basis for ADC variation observed in various tumors.