Purpose: Tumor heterogeneity is a hallmark of pancreatic ductal adenocarcinoma (PDAC). It determines tumor biology including tumor cellularity (i.e., amount of neoplastic cells and arrangement into clusters), which is related to the proliferative capacity and differentiation and the degree of desmoplasia among others. Given the close relation of tumor differentiation with differences in progression and therapy response or, e.g., the recently reported protective role of tumor stroma, we aimed at the noninvasive detection of PDAC groups, relevant for future personalized approaches. We hypothesized that histologic differences in PDAC tissue composition are detectable by the noninvasive diffusion weighted- (DW-) MRI-derived apparent diffusion coefficient (ADC) parameter.

Experimental design: PDAC cellularity was quantified histologically and correlated with the ADC parameter and survival in genetically engineered mouse models and human patients.

Results: Histologic analysis showed an inverse relationship of tumor cellularity and stroma content. Low tumor cellularity correlated with a significantly prolonged mean survival time (PDAClow = 21.93 months vs. PDACmed = 12.7 months; log-rank P < 0.001; HR = 2.23; CI, 1.41–3.53). Multivariate analysis using the Cox regression method confirmed tumor cellularity as an independent prognostic marker (P = 0.034; HR = 1.73; CI, 1.04–2.89). Tumor cellularity showed a strong negative correlation with the ADC parameter in murine (r = −0.84; CI, −0.90– −0.75) and human (r = −0.79; CI, −0.90 to −0.56) PDAC and high preoperative ADC values correlated with prolonged survival (ADChigh = 41.7 months; ADClow = 14.77 months; log rank, P = 0.040) in PDAC patients.

Conclusions: This study identifies high tumor cellularity as a negative prognostic factor in PDAC and supports the ADC parameter for the noninvasive identification of PDAC groups. Clin Cancer Res; 23(6); 1461–70. ©2016 AACR.

Translational Relevance

Pancreatic ductal adenocarcinoma (PDAC) is a virtually drug resistant and likely the most fatal of all malignant diseases. Better pretherapeutic detection of relevant prognostic and predictive groups would facilitate the development of personalized approaches.

We approach this problem in human and GEMM-based murine PDAC by thorough review of histology patterns and correlation with noninvasive diffusion weighted-MRI (DW-MRI). First, we define PDAC groups according to major differences in tumor cellularity and show their prognostic relevance in survival. Second, we identify the DW-MRI–derived apparent diffusion coefficient (ADC) as a candidate marker for the noninvasive estimation of PDAC cellularity. Finally, we show that the preoperative ADC parameter correlates with survival in PDAC patients regardless of histologic evaluation.

We believe that the ADC parameter is a strong candidate for the reliable distinction of PDAC cellularity groups irrespective of tumor resectability with great potential and clinically relevant impact on future therapeutic approaches in human PDAC.

Pancreatic adenocarcinoma (PDAC) is the most fatal cancer entity with rising incidence and long-term stagnant morbidity and mortality rates (1). Genetic and morphologic heterogeneity of human PDAC (hPDAC) complicates the identification and classification of groups for therapy individualization and arguably presents a key factor for the limited therapeutic advances. However, given the increasing armamentarium of chemotherapeutic and targeted options in the treatment of hPDAC, identification of key prognostic or predictive features of the underlying individual tumor biology may facilitate prospective evaluation in upcoming trials. Despite technical advances, no prognostic or predictive imaging biomarkers have been identified and implemented in clinical decision-making.

hPDAC very often exhibits strong desmoplasia, which is considered an effective barrier to drug delivery. However, recent evidence supports a model in which desmoplasia may be a protective factor with better prognosis (2). Furthermore, a significant and increased proportion of advanced hPDAC represents anaplastic and undifferentiated tumors with high cellularity and lack of abundant stroma and is characterized by noncohesive phenotypes showing a worse prognosis (3, 4). Molecular subtypes with more epithelial (classical) or mesenchymal (quasi-mesenchymal) features and different response to treatment have been proposed (5). The therapeutic relevance of these subtypes, which are among other features reflected by tumor cellularity, has recently been described in cell culture–based models (6).

Diffusion weighted-magnetic resonance imaging (DW-MRI) measures random motion of water molecules, expressed as an apparent diffusion coefficient (ADC), the weighted average of this coefficient within a given voxel (7). Intracellular diffusion is highly restricted due to structural barriers and this provides the basis for the detection of high cellularity voxel, for example, in oncologic imaging (8). Translation of this highly successful imaging technique to applications outside the brain has been partly hampered by technical challenges. However, consensus exists with regard to DW-MRI potential in particular in oncologic applications (9).

The aim of this study was to assess tumor cellularity defined as the amount of tumor cells in the specimen and their arrangement into clusters and stroma content in PDAC, hypothesizing that differences in tumor composition may translate into differences in patient survival. Further, we wanted to test the DW-MRI–derived ADC parameter as a surgery independent noninvasive biomarker, well-suited for the detection of structural differences in tissue composition.

This work has been performed and is reported in accordance with the reporting recommendations for tumor marker prognostic studies (REMARK; ref. 10).

Mouse strains

All experiments were performed according to the guidelines of the local Animal Use and Care Committee and are reported following the ARRIVE guidelines. Analyzed lesions were derived from previously described GEMM Ptf1awt/cre(C);Kraswt/LSL-KrasG12D(K)TP53(P)fl/wt, CKPR172H/wt, CKPfl/fl, CKEla-TGFa(T) (11–15). CKTPfl/wt and CKTPR172H/wt strains were bred from the above strains (Supplementary Fig. SF1A and SF1B).

MRI and data analysis in mice

Before all MRI experiments, mice were anesthetized by continuous gaseous infusion of 1.8% to 2% isoflurane (Abbott GmbH) for at least 10 minutes using a veterinary anesthesia System (Vetland Medical Sales and Services). During imaging, the dose was kept at 2% isoflurane, animal temperature was maintained and continuously monitored and eyes were protected with an eye ointment.

Mice were imaged in prone position using a 47-mm microscopy surface coil inside the clinical 1.5 T MRI System (Achieva 1.5T, Philips Medical Systems). For tumor detection, an axial multi-slice T2-weighted (T2w) TSE sequence (resolution 0.3 × 0.3 × 0.7 mm3, minimum 25 slices, TE = 90 milliseconds, TR > 3 seconds) was applied on the mouse abdomen during free breathing. Following, an axial multislice diffusion-weighted MRI (DW-MRI) sequence covering the tumors was performed (resolution 0.7 × 0.7 × 1.5 mm3, EPI factor = 45, TR/TE = 3,000/60 milliseconds, b0–2 values = 20, 200, 600 s/mm2, averages = 10).

MRI data analysis was performed using Osirix Imaging Software and in-house software written in IDL (ITT). ROI size ranged from 2 to 30 mm2 and no correlation was noted between size of ROI and PDAC group (r = 0.21; CI, 0.006–0.4). Histologically confirmed areas of necrosis were excluded from all imaging analyses. Imaging and corresponding histology ROIs were correlated slice based for every tumor region. All ROIs were manually defined based on histologically defined tumor borders and quality of imaging data. In animals with several slices of the same tumor subtype, the slice with the lowest potential for partial volume (i.e., central slice) was taken. Up to three tumors per animal (i.e., head, neck and/or tail region) were included. A total number of 68 animals and 100 lesions were included into the study. All numbers are given as number of lesions used in particular study.

Patients

Written informed consent was obtained from all patients. The Institutional Review Board (IRB) approved prospective data collection and review of the patient charts for this project. Analysis was conducted on a pseudonymized data set.

In this exploratory retrospective study, a cohort of 123 patients that had undergone an elective pancreatic resection with a final histopathologic diagnosis of hPDAC between July 2007 and December 2014 was used. The complete clinical follow up was used until May 2015 (time of the analysis). Depending on the analyzed issue, this cohort was divided into 3 subcohorts. An overview of each subcohort used in this study is provided in Table 1 (Supplementary Tables S1–S3 and Supplementary Figs. S2–S4). Available formalin-fixed and paraffin-embedded surgical hPDAC specimens and pathology reports were obtained from the Institute of Pathology of the Technische Universität München, Germany. For image analysis, only 1.5 Tesla DW-MRI performed at the Institute of Radiology of the Technische Universität München prior to surgery was taken into account. For correlation of ADC value and histology (subcohort 1), a total of 52 patients that had received preoperative DW-MRI were closely reviewed and 21 patients with exact regional correlation were included (Table 1, Supplementary Table ST1, and Supplementary Fig. SF2). In retrospectively reviewed specimen, pathologists and radiologists together reviewed the pathologic reports and identified the anatomical landmarks. Only cases in which a proper co-registration of the pathologic area and ADC map was possible were included in the analysis of subcohort 1. For survival analysis based on histologic findings (subcohort 2), a total of 123 patients were reviewed and 96 patients were included (Table 1, Supplementary Table ST2, and Supplementary Fig. SF3). Median follow-up for patients of subcohort 2 was 17.73 months (2.5–82.03 months). The respective PDAC group (hPDAClow, hPDACmed, hPDAChigh) was estimated from all available archived histologic slides (at least 3 different localizations per tumor). For survival analysis based on lowest mean ADC value (subcohort 3), a total of 52 patients that had received preoperative DW-MRI was reviewed and 44 patients with high-quality ADC maps throughout the entire tumor were included (Table 1, Supplementary Table ST3, and Supplementary Fig. SF4). Median follow-up for patients of subcohort 3 was 28.43 months (2.63–82.13 months).

Table 1.

Summary of surgical subcohorts used in the study.

Surgical subcohortNumber of patients analyzedPurposeParameterStatistical method
N=21 (thereof 4 patients with 2 lesions, i.e. 25 lesions) to correlateregional tumor histopathology and pre-operative mean ADC value Histology
  • analysis of the percentage tumor cells, stroma and open spaces

ADC value
  • manual segmentation of corresponding mean ADC value (>10 mm diameter ROI)

  • see Suppl. Fig. 2, Suppl. Tab 1

 
Pearson's correlation analysis for each histological parameter and regional mean ADC value 2-group comparison (unpaired t test with Welch's correction (**), Mann Whitney test (*)) 
N=96 to test dependence of histological tumor subtype and survival Histopathological subtype
  • PDACmed (tubular growth pattern and clusters of at least 1 mm diameter with small irregular glands and 30–70% of tumor cells within the cluster area)

  • PDAClow (no tumor cell cluster greater equal 1 mm2 and abundant desmoplasia)

Survival time of patients included in the study
  • see Suppl. Fig. 3, Suppl. Tab 2

 
Kaplan-Meier survival estimates for PDAClow and PDACmed (Log-Rank test)Multivariate Cox regression model analysis 
N=44 to test dependence of lowest pre-operative ADC value and survival Regional ADC value
  • Manual segmentation of lowest mean ADC value of entire tumor (>10 mm diameter ROI)

Survival time of patients included in the study
  • see Suppl. Fig. 4, Suppl. Tab. 3

 
Kaplan-Meier survival estimates for ADClow and ADChigh (Log-Rank test) and Maximally Selected Rank Statistics test estimating the cut of for 2 groups with unknown relationship: ADClow and ADChigh 
Surgical subcohortNumber of patients analyzedPurposeParameterStatistical method
N=21 (thereof 4 patients with 2 lesions, i.e. 25 lesions) to correlateregional tumor histopathology and pre-operative mean ADC value Histology
  • analysis of the percentage tumor cells, stroma and open spaces

ADC value
  • manual segmentation of corresponding mean ADC value (>10 mm diameter ROI)

  • see Suppl. Fig. 2, Suppl. Tab 1

 
Pearson's correlation analysis for each histological parameter and regional mean ADC value 2-group comparison (unpaired t test with Welch's correction (**), Mann Whitney test (*)) 
N=96 to test dependence of histological tumor subtype and survival Histopathological subtype
  • PDACmed (tubular growth pattern and clusters of at least 1 mm diameter with small irregular glands and 30–70% of tumor cells within the cluster area)

  • PDAClow (no tumor cell cluster greater equal 1 mm2 and abundant desmoplasia)

Survival time of patients included in the study
  • see Suppl. Fig. 3, Suppl. Tab 2

 
Kaplan-Meier survival estimates for PDAClow and PDACmed (Log-Rank test)Multivariate Cox regression model analysis 
N=44 to test dependence of lowest pre-operative ADC value and survival Regional ADC value
  • Manual segmentation of lowest mean ADC value of entire tumor (>10 mm diameter ROI)

Survival time of patients included in the study
  • see Suppl. Fig. 4, Suppl. Tab. 3

 
Kaplan-Meier survival estimates for ADClow and ADChigh (Log-Rank test) and Maximally Selected Rank Statistics test estimating the cut of for 2 groups with unknown relationship: ADClow and ADChigh 

MRI and imaging data analysis in humans

MRI was performed on a 1.5-T system (Magnetom Avanto, Siemens Medical Solutions) with two six-channel body-phased array coils anterior and two spine clusters (three channels each) posterior. In addition to the diffusion-weighted sequences, at least a coronal T2-weighted half-Fourier single-shot turbo spin-echo (HASTE) sequence and an axial T2-weighted turbo spin-echo sequence were acquired. Diffusion-weighted images were acquired using a single-shot echo-planar imaging sequence. To acquire images with a high contrast-to-noise ratio for optimal conspicuity of pancreas tumors while keeping the influence of “pseudo-diffusion” by means of perfusion effects low, the minimum gradient factor (b value) was set at 50 s/mm2. Thus, the gradient factors (b values) were 50, 300, and 600 s/mm2. The technical parameters were as follows: echo time, 69 milliseconds; echo train length, 58; echo spacing, 0.69 milliseconds; receiver bandwidth, 1,736 Hz/pixel; spectral fat saturation; field of view, 263 × 350 mm; matrix, 144 × 192; section thickness, 5 mm. For shortening of the echo train length, integrated parallel imaging techniques (iPAT) by means of generalized auto-calibrating partially parallel acquisitions (GRAPPA) with a twofold acceleration factor were used. For respiratory triggering, prospective acquisition correction (PACE) was implemented. Data were acquired during the end-expiratory phase. Both experienced radiologist (RB) and pathologists (IE and KS) correlated tumor specimen histopathology and ADC map for lesion analysis (subcohort 1; Table 1, Supplementary Table S1, and Supplementary Fig. SF2). Corresponding spherical ROIs with a minimal diameter of greater equal 10 mm were placed in tumor of interest. For the ADC-based survival analysis (subcohort 3; Table 1, Supplementary Table ST3, and Supplementary Fig. SF4), two experienced radiologists (RB, IH) consensually placed the ROIs with a minimal diameter of greater equal 10 mm into the region of the lowest ADC of the entire tumor blinded to histopathologic findings and clinical outcome. The regional ADC values were recorded as a mean value with standard deviation.

Histologic data analysis

Upon euthanization, mPDAC were removed from the abdominal cavity with adjacent organ structures (liver, spleen, gut, kidneys) to prevent change in orientation and facilitate exact correlation of imaging plane and histology based on additional anatomical landmarks. After formalin fixation and paraffin embedding and hematoxylin and eosin (H&E) staining, axial histologic slices through mouse abdomen with 400 μm distance were correlated with DW-MRI. Stromal content and amount of tumor cells and the percentage of open spaces (ducts, cysts, vessels) were determined using H&E- and Movat's pentachrome (modified according to Verhoeff, Morphisto GmbH) stained slides by two experienced pathologist (IE and KS) blinded for imaging findings and mouse genotype. The tumors were reviewed at least three times and the results were reproducible in all three replicates.

Grading of mPDAC and hPDAC was performed according to WHO classification of tumors of the pancreas considering actual consensus reports of pathology of GEMM (16, 17). Tumors that revealed a heterogeneous phenotype with partially well and poorly differentiated morphology were always classified to the worse grade corresponding to clinical standards.

Definition of histopathologic groups

Tumor cellularity has been defined as the amount of tumor cells in the specimen and their arrangement into clusters. Based on the PDAC groups identified after co-registration of histology with DW-MRI in subcohort 1, a definition of hPDAC classification based on the occurrence of tumor cell clusters greater equal 1 mm2 was developed and subsequently also applied to subcohort 2. hPDAC were classified as high-level cellular (hPDAChigh), in case of predominantly solid growth pattern with >70% tumor cells (reflecting PDAC variants such as adenosquamous or medullary carcinoma) and only very small amounts of accompanying stroma within the cluster region (Supplementary Fig. SF5A). Medium-level cellular tumors (hPDACmed) showed a classical tubular growth pattern but exhibited clusters of small irregular glands (30–70% of tumor cells within the cluster area). Morphologically, these clusters consisted of neoplastic cells with solid, cribriform, or gyriform growth patterns or with single pleomorphic tumor cells (Supplementary Fig. SF5B). PDAC were classified as low-level cellular (hPDAClow) if no tumor cell cluster greater equal 1 mm2 was found in any of the localizations investigated (Supplementary Fig. SF5C). Classification of hPDAC was performed by two experienced pathologists (IE, KS). The tumors were reviewed blinded at least three times and the results were reproducible in all three replicates.

As for mPDAC, we applied the same classification. Cellularity cut offs differed due to the higher amount of invasive cancer cells observed in the murine specimens in correlation with DW-MRI: 1% to 40% for mPDAClow, 40% to 85% for mPDACmed and 85% to 100% for mPDAChigh.

Statistical analysis

No formal sample size calculation was performed prior to the study. Based on the actual sample size, a true correlation of murine histology parameter and ADC value (r = 0.3) and a true hazard ratio between the human survival groups was detected with a power of 80% and 90%, respectively. Data are illustrated in dot plots stratified for groups. Most statistical analyses were done in GraphPad Prism version 6.0e. For all statistical tests, a level of significance of 5% was used. ANOVA (**) was performed for group comparisons of normally distributed quantities, if data were observed to be skewed, the Kruskal–Wallis (*) test was conducted. For pairwise group comparisons, Holm–Sidak's (**) or Dunn's post hoc test (*) was applied, respectively. For two groups comparison, either unpaired t test with Welch's correction (**) or Mann–Whitney test (*) was used. P values were always used from two-sided calculation. The strength of association between quantitative measures was assessed by Pearson's correlation coefficient (all data sets revealed normal distribution) and a corresponding 95% confidence interval is presented.

Survival analysis was performed using the Kaplan–Meier method; differences were evaluated with the log-rank test. In addition, Cox regression was performed to estimate hazard ratios with 95% confidence intervals (SSPS version 23). For human samples, analysis overall survival (OS) was defined as time from resection (subcohort 2) or as time from preoperative imaging (subcohort 3) until death after discharge or until last follow-up collected till May 2015. In addition, for subcohort 2, a multivariable Cox regression model was fit to the data including the covariates PDAC (med vs. low), age, sex, T-status (T), lymph node infiltration (N), grading (G), resection margin (R) and CTX to assess whether PDAC is associated with overall survival after adjustment for these predictors. To determine a cut-off value for lowest ADC value (subcohort 3) to best discriminate high-risk from low-risk patients with regard to overall survival, the value providing the highest log-rank statistic was used. A P value for association between the lowest ADC value and overall survival was derived by a permutation tests, as proposed previously (18), using the statistical software R (19) and its package coin (20). A multivariable regression model could not be fitted to the subcohort 3 data due to the small number of observed events (21). Therefore, the permutation test assessing the association between overall survival and lowest ADC was also performed stratified by each of the relevant categorical variables (T, N, G, R, CTX).

Tumor cellularity reciprocally correlates with stroma content and is a prognostic factor of survival in human PDAC

High inter- and intratumoral genetic and molecular heterogeneity have been identified as potential obstacles to successful treatment of human PDAC (hPDAC; refs. 22, 23). Besides rather homogenous alterations in KRAS, TP53, CDKN2A and SMAD4, other genetic alterations are highly heterogeneous and go along with metabolic and morphologic heterogeneity, which likely adds to the ongoing low therapeutic benefits in human PDAC (hPDAC; refs. 5, 23–25). This heterogeneity is also reflected by intratumoral differences in tumor architecture, that is, the amount of tumor cell, their arrangement into clusters and stroma (Fig. 1A). We analyzed regional tumor cellularity and stromal content in a cohort of hPDAC specimen from patients that underwent an elective pancreatic resection at the Technical University of Munich (Table 1, Supplementary Table ST1, and Supplementary Fig. SF2) and found an inverse relationship of the two tissue components (r = −0.93; CI, −0.97 to −0.86). We next analyzed and classified hPDAC specimen (Table 1, Supplementary Table ST2, and Supplementary Fig. SF3) according to differentiation grade, neoplastic cellularity, and stromal content (16, 17). Based on tumor cellularity we identified three groups: (i) low overall tumor cellularity, gland formation, and abundant desmoplastic stroma (hPDAClow), (ii) occurrence of cancer cell clusters with intermediate cellularity or parts of solid growth pattern (hPDACmed), and (iii) high cellularity and low stroma content (hPDAChigh) as shown by H&E and Movat staining (Fig. 1B and Supplementary Fig. SF5). To test the prognostic value of this proposed tumor stratification, we performed a survival analysis of hPDACmed versus hPDAClow and found a significantly worse survival of hPDACmed patients (12.7 months vs. 20.93 months; P = 0.001; HR = 2.23; CI, 1.41–3.54; Fig. 1C). We excluded hPDAChigh patients from the survival analysis because of the low frequency of this group in our retrospective surgical cohort. In a multivariable regression model adjusting for age, sex, pT, pN, grading, R, and CTX a significantly higher risk was estimated for hPDACmed patients compared to hPDAClow patients (P = 0.034; HR = 1.73; CI, 1.04–2.89; Supplementary Table ST4). As expected, higher grading also resulted in worse median survival time (G3 = 13.2 vs. G2 = 23.13 months; P < 0.0001; HR = 2.80; CI, 1.76–4.45; Fig. 1D). Interestingly, classification based on tumor cellularity and grading identified a subgroup of G2 tumors of intermediate cellularity (hPDACmed/G2) with significantly worse prognosis (16.17 months; P = 0.001 compared to hPDACmed/G3 34.4 months; HR = 4.85; CI, 1.64–7.19; Fig. 1E, black line). Based on these findings, tumor cellularity is a potential prognostic factor of survival in hPDAC.

Figure 1.

Histologic heterogeneity of human PDAC correlates with survival prognosis. A, Representative large-scale overview and higher magnification micrographs of hPDAC reveal regional differences in tumor cellularity and stromal content, that is, hPDAClow and hPDACmed. H&E staining, scale baroverview 1 cm, scale barmagnification 200 μm. B, Classification of human histology according to tumor cellularity. Exemplary high magnification photomicrographs of H&E and Movat stained human hPDAClow, hPDACmed, and hPDAChigh. Scale bar 100 μm. C and D, Kaplan–Meier survival analysis of hPDAClow (n = 61, 20.93 months) and hPDACmed (n = 35, 12.7 months) as well as G2 (n = 47, 23.13 months) and G3 (n = 49, 13.2 months) hPDAC patients (subcohort 2). E, Kaplan–Meier survival analysis taking both grading and tumor into account (hPDAClow/G2, n = 32, 34.4 months; hPDAClow/G3+4, n = 29, 14.6 months; hPDACmed/G2, n = 15, 16.2 months; hPDACmed/G3+4, n = 20, 11.6 months). All groups were compared using log-rank test.

Figure 1.

Histologic heterogeneity of human PDAC correlates with survival prognosis. A, Representative large-scale overview and higher magnification micrographs of hPDAC reveal regional differences in tumor cellularity and stromal content, that is, hPDAClow and hPDACmed. H&E staining, scale baroverview 1 cm, scale barmagnification 200 μm. B, Classification of human histology according to tumor cellularity. Exemplary high magnification photomicrographs of H&E and Movat stained human hPDAClow, hPDACmed, and hPDAChigh. Scale bar 100 μm. C and D, Kaplan–Meier survival analysis of hPDAClow (n = 61, 20.93 months) and hPDACmed (n = 35, 12.7 months) as well as G2 (n = 47, 23.13 months) and G3 (n = 49, 13.2 months) hPDAC patients (subcohort 2). E, Kaplan–Meier survival analysis taking both grading and tumor into account (hPDAClow/G2, n = 32, 34.4 months; hPDAClow/G3+4, n = 29, 14.6 months; hPDACmed/G2, n = 15, 16.2 months; hPDACmed/G3+4, n = 20, 11.6 months). All groups were compared using log-rank test.

Close modal

DW-MRI reliably detects tumor cellularity in murine PDAC

In contrast to histologic grading, tumor cellularity and stroma content present structural tissue parameters, potentially quantifiable by noninvasive imaging using DW-MRI (8). To verify the DW-MRI–derived ADC parameter as a potential imaging biomarker of tumor cellularity in murine PDAC (mPDAC), we next employed different KrasG12D-based genetically engineered mouse models (GEMM) of endogenous mPDAC to mirror the morphologic inter- and intratumoral heterogeneity of hPDAC. Tumors derived from six different genotypes were classified histologically according to the identified hPDAC groups (Fig. 2A and Supplementary Fig. SF1A). Here, we also found an inverse relationship of tumor cellularity and stroma (r = –0.61; CI, –0.72 to –0.47). We next established a platform for the slice-based regional correlation of histologies and ADC maps with respective ADC histograms as exemplarily shown for the different mPDAC (Fig. 2B–D). To further validate the regional ADC value as a marker of tumor cellularity, we performed a semiquantitative correlation. Here, the ADC parameter showed a strong negative correlation with tumor cellularity (r = −0.84; CI = −0.90 to −0.75; Fig. 3A), a strong positive correlation with tumor stroma (r = 0.77; CI, 0.63–0.86; Fig. 3B) and a moderate positive correlation with cystic and ductal spaces (r = 0.60; CI = –0.40 to 0.75; Fig. 3C). As expected, we found major variability between individual mPDAC regions in agreement with high tissue heterogeneity. However, an excellent distinction and little overlap was evident for ADC values of mPDAClow (1.03 ± 0.09), mPDACmed (0.82 ± 0.06), and mPDAChigh (0.68 ± 0.06; Fig. 3D). In contrast to tumor grading (Fig. 3E), which could not be clearly distinguished by ADC (G2 = 0.,94 ± 0.14, G3 = 0.93 ± 0.16, G4 = 0.73 ± 0.13). Because of genotype-related differences in tumor onset, no survival analyses were performed for mice (Supplementary Fig. SF1B).

Figure 2.

The ADC parameter predicts histologic heterogeneity of murine PDAC with high sensitivity and specificity. A, mPDAC faithfully recapitulates heterogeneity of hPDAC. Exemplary high magnification photomicrographs of H&E and Movat stained murine lesions of mPDAClow, mPDACmed, and mPDAChigh tumor. Scale bar 100 μm. B–D, Slice-based correlation of imaging and histology. T2w image, ADC map, and corresponding whole pancreas (yellow line) as well as high magnification (box, scale bar = 50 μm) micrographs representing heterogeneity in mPDAC. Each region of mPDAClow (fine dotted line), mPDACmed (rough dotted line), and mPDAChigh (solid line) is well correlated and clearly distinguishable on ADC map. Corresponding histogram analyses of different regions are shown below.

Figure 2.

The ADC parameter predicts histologic heterogeneity of murine PDAC with high sensitivity and specificity. A, mPDAC faithfully recapitulates heterogeneity of hPDAC. Exemplary high magnification photomicrographs of H&E and Movat stained murine lesions of mPDAClow, mPDACmed, and mPDAChigh tumor. Scale bar 100 μm. B–D, Slice-based correlation of imaging and histology. T2w image, ADC map, and corresponding whole pancreas (yellow line) as well as high magnification (box, scale bar = 50 μm) micrographs representing heterogeneity in mPDAC. Each region of mPDAClow (fine dotted line), mPDACmed (rough dotted line), and mPDAChigh (solid line) is well correlated and clearly distinguishable on ADC map. Corresponding histogram analyses of different regions are shown below.

Close modal
Figure 3.

The ADC parameter predicts histologic heterogeneity, and stroma and tumor cell content of murine PDAC. A and B, In mPDAC, ADC correlates well with tumor cellularity (n = 59) and stromal content (n = 52), and (C) moderately with open spaces such as ducts and cysts (n = 53). All scatterplots are presented with regression line, corresponding Pearson's correlation coefficient (r) and confidence interval (CI). D, Mean ADC values of mPDAClow (n = 49), mPDACmed (n = 38), and mPDAChigh (n = 13) groups differ significantly and exhibit little overlap (**), in contrast to (E) mPDACs of different grading (G1+2, n = 13; G3, n = 50; G4, n = 11; **).

Figure 3.

The ADC parameter predicts histologic heterogeneity, and stroma and tumor cell content of murine PDAC. A and B, In mPDAC, ADC correlates well with tumor cellularity (n = 59) and stromal content (n = 52), and (C) moderately with open spaces such as ducts and cysts (n = 53). All scatterplots are presented with regression line, corresponding Pearson's correlation coefficient (r) and confidence interval (CI). D, Mean ADC values of mPDAClow (n = 49), mPDACmed (n = 38), and mPDAChigh (n = 13) groups differ significantly and exhibit little overlap (**), in contrast to (E) mPDACs of different grading (G1+2, n = 13; G3, n = 50; G4, n = 11; **).

Close modal

The ADC value is a noninvasive marker of tumor cellularity and correlates with survival in human PDAC

Encouraged by these results, we retrospectively matched regional ADC values and histopathology (subcohort 1) that also received DW-MRI prior to tumor resection (Fig. 4A, Table 1, Supplementary Table ST1, and Supplementary Fig. ST2). In agreement with the results obtained for mPDAC, a negative correlation of the ADC parameter with tumor cellularity (r = –0.79; CI, –0.90 to –0.56; Fig. 4B) and a positive correlation with tumor stroma (r = 0.77; CI, 0.54–0.89; Fig. 4C) was observed. No apparent correlation was evident for open spaces (r = 0.21; CI, –0.19 to 0.55). Nevertheless, hPDAClow and hPDACmed could be clearly distinguished (1.69 ± 0.19 vs. 1.19 ± 0.18; Fig. 4D) with little overlap between them. As in mice, ADC values of tumors with different histopathologic gradings revealed major overlap (G2 = 1.57 ± 0.29, G3 = 1.33 ± 0.36, G4 = 1.02 ± 0.03; Fig. 4E), precluding noninvasive differentiation.

Figure 4.

ADC correlates with tumor cellularity and stromal content in human PDAC. A, Representative MR images depicting T2w images, ADC maps, and micrographs of H&E staining from corresponding regions of the same tumor in one patient, exhibiting hPDAClow and hPDACmed phenotypes. Scale bar 100 μm. B and C, ADC values strongly correlate with tumor cellularity and stromal content (n = 25 lesions, subcohort 1). All scatterplots are presented with regression line, corresponding Pearson's correlation coefficient (r) and confidence interval (CI). D, Mean ADC values of hPDAClow (n = 14) and hPDACmed (n = 11) groups differ significantly and groups exhibit little overlap (**), whereas (E) hPDACs of different grading (G2, n = 16; G3, n = 6; G4, n = 2) again show major overlap (*).

Figure 4.

ADC correlates with tumor cellularity and stromal content in human PDAC. A, Representative MR images depicting T2w images, ADC maps, and micrographs of H&E staining from corresponding regions of the same tumor in one patient, exhibiting hPDAClow and hPDACmed phenotypes. Scale bar 100 μm. B and C, ADC values strongly correlate with tumor cellularity and stromal content (n = 25 lesions, subcohort 1). All scatterplots are presented with regression line, corresponding Pearson's correlation coefficient (r) and confidence interval (CI). D, Mean ADC values of hPDAClow (n = 14) and hPDACmed (n = 11) groups differ significantly and groups exhibit little overlap (**), whereas (E) hPDACs of different grading (G2, n = 16; G3, n = 6; G4, n = 2) again show major overlap (*).

Close modal

To test the clinical potential of the ADC parameter, we next performed a retrospective Kaplan–Meier analysis (Table 1, Supplementary Table ST3, and Supplementary Fig. SF4), with a mean lowest regional ADC value of 1.27 of all tumors analyzed. ADClow (≤1.27) tumors revealed significantly shorter median survival compared to ADChigh (>1.27) patients (14.77 months vs. 41.70 months, log-rank P = 0.040; Fig. 5A).

Figure 5.

Regional lowest ADC value is predictive for survival in PDAC patients. A, Kaplan–Meier survival analysis of ADClow (≤1.27, n = 20, 14.8 months) and ADChigh (>1.27, n = 24, 41.7 months) hPDAC patients (subcohort 3). B, Kaplan–Meier survival analysis of ADClow (≤1.08, n = 8, 8.25 months) and ADChigh (>1.08, n = 36, 29.5 months) hPDAC patients (subcohort 3). The cut-off was determined by the maximally selected rank statistics test. C, Schematic representation of restricted diffusion (ADClow) in tissues exhibiting high tumor celluarity (PDAChigh). A smaller slope of the curve (red line), fitted through different b-value measurements of the same tissue, translates into a lower ADC value determining PDAChigh phenotype.

Figure 5.

Regional lowest ADC value is predictive for survival in PDAC patients. A, Kaplan–Meier survival analysis of ADClow (≤1.27, n = 20, 14.8 months) and ADChigh (>1.27, n = 24, 41.7 months) hPDAC patients (subcohort 3). B, Kaplan–Meier survival analysis of ADClow (≤1.08, n = 8, 8.25 months) and ADChigh (>1.08, n = 36, 29.5 months) hPDAC patients (subcohort 3). The cut-off was determined by the maximally selected rank statistics test. C, Schematic representation of restricted diffusion (ADClow) in tissues exhibiting high tumor celluarity (PDAChigh). A smaller slope of the curve (red line), fitted through different b-value measurements of the same tissue, translates into a lower ADC value determining PDAChigh phenotype.

Close modal

To determine a cut-off value for lowest ADC to best discriminate high-risk from low-risk patients with regard to overall survival, we applied the maximally selected rank statistics as recommended by Altman and colleagues (10). This test estimates the cut of two groups with unknown relationship for continuous variables. As shown in Fig. 5B, this analysis identified two groups of different survival that are separated by an ADC cut off value of 1.08 and that exhibits an even worse prognosis (8.25 months vs. 29.47 months). Because of the low number of events, no statistical significance could be reached. Similar results were also found in the analyses stratified for relevant confounders. Each of the stratified analyses revealed a cutoff value of 1.08 (Supplementary Table ST5).

In this work, we identify tumor cellularity, i.e., the amount of tumor cell in the specimen and their arrangement into clusters as a negative prognostic factor for overall survival in patients resected for PDAC. By systematically correlating tumor cellularity with the DW-MRI derived ADC parameter we show its high potential as a non-invasive marker for PDAC group identification and patient stratification regardless of the tumor resectability.

Tumor cellularity likely presents a marker of therapeutic relevance. In comparison to surgical specimen used in our study, which are more reflective of early stage tumors, autopsy studies show a higher prevalence of highly cellular and non-cohesive tumors (approx. 20% vs. < 1%), possibly at least in part the result of a later appearance in a step-wise tumor progression (4). Thus, considering drug-response and inherent or acquired therapy resistance, non-invasive identification of tumor cellularity before and during therapy could directly impact the clinical decision process and facilitate evaluation of guided therapeutic approaches. Mounting evidence suggests a pivotal role of tumor stroma in sustaining and driving cancer cell proliferation in hPDAC. Preclinical studies targeting tumor stroma (13, 14, 26) have shown improved drug delivery and prolonged survival. However, despite these promising results, two clinical trials evaluating sonic hedgehog inhibition mediated stroma depletion were stopped due to poor efficacy, possibly due to an additional restraining role of stroma in PDAC progression (27). Our histopathologic analyses suggest a better prognosis for tumors exhibiting high stroma content. Along this line, a recent study revealed a positive effect of tumor stroma on OS and RFS in a cohort of curatively resected hPDAC (28). Furthermore, in a study on neoadjuvant chemoradiation, patients exhibiting higher stroma content (and higher perfusion values) exhibited better response and outcome despite an inverse correlation of tumor stroma and gemcitabine delivery (29). Whereas the authors suggested arterial-venous shunting as a possible explanation for this observation, others and our data would argue for a lower tumor cell load as one potential reason for increased survival in this cohort.

In hPDAC, imaging has been of limited prognostic or predictive value (30–32) and thus has not been implemented in clinical decision-making. Based on our findings we propose the ADC parameter, an imaging parameter sensitive to tissue structure, as a highly promising candidate for noninvasive identification and monitoring of tumor cellularity in hPDAC before and during therapy. High cellularity accompanied by low stromal content in hPDAChigh tumors results in stronger restriction of water movement in the tissue reflected by higher measured b and subsequently lower calculated ADC values (Fig. 5C). In support of our observations, a recent study identified low baseline ADC as a negative predictor of survival and therapy response, respectively (33). Previously published contradicting results for the correlation of the ADC value and tumor histopathology (34–36) may at least in part be explained by technical differences in image acquisition and ADC calculation (37–39). Accordingly, the implementation of standardized protocols has been put forward by an expert panel as a requisite for fast clinical translation of this technique (9).

In summary, the presented work supports the clinical relevance of differentiating PDAC based on tumor cellularity and reveals a high sensitivity and specificity of the ADC parameter for noninvasive detection of PDAC groups. We therefore propose DW-MRI as a fast, radiation and contrast agent-free imaging method, easily integrated into routine clinical PDAC imaging protocols. Reliable, noninvasive assessment of tumor cellularity of a particular tumor by means of ADC calculation may facilitate stratification of PDAC groups for outcome analysis and personalized preclinical and clinical therapeutic intervention trials.

M. Schwaiger reports receiving commercial research grants from Siemens Medical Solution. No potential conflicts of interest were disclosed by the other authors.

Conception and design: I. Heid, M. Trajkovic-Arsic, M. Settles, J. Kleeff, E.J. Rummeny, I. Esposito, J.T. Siveke, R.F. Braren

Development of methodology: I. Heid, K. Steiger, M. Settles, A. Steingötter, I. Esposito, J.T. Siveke, R.F. Braren

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I. Heid, K. Steiger, M. Trajkovic-Arsic, M. Settles, M.R. Eßwein, M. Erkan, J. Kleeff, C. Jäger, H. Friess, R.M. Schmid, I. Esposito, J.T. Siveke

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I. Heid, K. Steiger, M. Trajkovic-Arsic, M. Settles, M.R. Eßwein, M. Erkan, C. Jäger, B. Haller, A. Steingötter, R.M. Schmid, E.J. Rummeny, I. Esposito, J.T. Siveke, R.F. Braren

Writing, review, and/or revision of the manuscript: I. Heid, K. Steiger, M. Trajkovic-Arsic, M. Settles, M. Erkan, J. Kleeff, A. Steingötter, M. Schwaiger, I. Esposito, J.T. Siveke, R.F. Braren

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): I. Heid, M. Settles, C. Jäger, H. Friess, R.M. Schmid, E.J. Rummeny

Study supervision: I. Heid, E.J. Rummeny, I. Esposito, J.T. Siveke, R.F. Braren

We would like to thank Dr. A. Berns, Dr. E. Sandgren, Dr. D. Tuveson, and Dr. T. Jacks for providing transgenic animals. We thank Tamara Schilling, Juliane Blumenberg for excellent help with murine MR imaging and anesthesia; Mathilde Neuhofer and Iryna Skuratovska for expert histopathologic data preparation.

This work was supported by the German Research Foundation (DFG) within the SFB-Initiative 824 (collaborative research center), “Imaging for Selection, Monitoring and Individualization of Cancer Therapies” (SFB824, project C4, C6, and Z2), the European Community's Seventh Framework Program (FP7/CAM-PaC) under grant agreement no. 602783, and by the German Cancer Consortium (DKTK; to M. Schwaiger, R.M. Schmid, and J.T. Siveke).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Raimondi
S
,
Maisonneuve
P
,
Lowenfels
AB
. 
Epidemiology of pancreatic cancer: an overview
.
Nat Rev Gastroenterol Hepatol
2009
;
6
:
699
708
.
2.
Ozdemir
BC
,
Pentcheva-Hoang
T
,
Carstens
JL
,
Zheng
X
,
Wu
CC
,
Simpson
TR
, et al
Depletion of carcinoma-associated fibroblasts and fibrosis induces immunosuppression and accelerates pancreas cancer with reduced survival
.
Cancer Cell
2014
;
25
:
719
34
.
3.
Polyak
K
,
Weinberg
RA
. 
Transitions between epithelial and mesenchymal states: acquisition of malignant and stem cell traits
.
Nat Rev Cancer
2009
;
9
:
265
73
.
4.
Winter
JM
,
Ting
AH
,
Vilardell
F
,
Gallmeier
E
,
Baylin
SB
,
Hruban
RH
, et al
Absence of E-cadherin expression distinguishes noncohesive from cohesive pancreatic cancer
.
Clin Cancer Res
2008
;
14
:
412
8
.
5.
Collisson
EA
,
Sadanandam
A
,
Olson
P
,
Gibb
WJ
,
Truitt
M
,
Gu
S
, et al
Subtypes of pancreatic ductal adenocarcinoma and their differing responses to therapy
.
Nat Med
2011
;
17
:
500
3
.
6.
Mirzoeva
OK
,
Collisson
EA
,
Schaefer
PM
,
Hann
B
,
Hom
YK
,
Ko
AH
, et al
Subtype-specific MEK-PI3 kinase feedback as a therapeutic target in pancreatic adenocarcinoma
.
Mol Cancer Ther
2013
;
12
:
2213
25
.
7.
Bammer
R
. 
Basic principles of diffusion-weighted imaging
.
Eur J Radiol
2003
;
45
:
169
84
.
8.
Koh
DM
,
Collins
DJ
. 
Diffusion-weighted MRI in the body: applications and challenges in oncology
.
AJR Am J Roentgenol
2007
;
188
:
1622
35
.
9.
Taouli
B
,
Beer
AJ
,
Chenevert
T
,
Collins
D
,
Lehman
C
,
Matos
C
, et al
Diffusion-weighted imaging outside the brain: consensus statement from an ISMRM-sponsored workshop
.
J Magn Reson Imaging
2016
;
44
:
521
40
.
10.
Altman
DG
,
McShane
LM
,
Sauerbrei
W
,
Taube
SE
. 
Reporting recommendations for tumor marker prognostic studies (REMARK): explanation and elaboration
.
PLoS Med
2012
;
9
:
e1001216
.
11.
Ardito
CM
,
Gruner
BM
,
Takeuchi
KK
,
Lubeseder-Martellato
C
,
Teichmann
N
,
Mazur
PK
, et al
EGF receptor is required for KRAS-induced pancreatic tumorigenesis
.
Cancer Cell
2012
;
22
:
304
17
.
12.
Hingorani
SR
,
Wang
L
,
Multani
AS
,
Combs
C
,
Deramaudt
TB
,
Hruban
RH
, et al
Trp53R172H and KrasG12D cooperate to promote chromosomal instability and widely metastatic pancreatic ductal adenocarcinoma in mice
.
Cancer Cell
2005
;
7
:
469
83
.
13.
Olive
KP
,
Jacobetz
MA
,
Davidson
CJ
,
Gopinathan
A
,
McIntyre
D
,
Honess
D
, et al
Inhibition of Hedgehog signaling enhances delivery of chemotherapy in a mouse model of pancreatic cancer
.
Science
2009
;
324
:
1457
61
.
14.
Provenzano
PP
,
Cuevas
C
,
Chang
AE
,
Goel
VK
,
Von Hoff
DD
,
Hingorani
SR
. 
Enzymatic targeting of the stroma ablates physical barriers to treatment of pancreatic ductal adenocarcinoma
.
Cancer Cell
2012
;
21
:
418
29
.
15.
Siveke
JT
,
Einwachter
H
,
Sipos
B
,
Lubeseder-Martellato
C
,
Kloppel
G
,
Schmid
RM
. 
Concomitant pancreatic activation of Kras(G12D) and Tgfa results in cystic papillary neoplasms reminiscent of human IPMN
.
Cancer Cell
2007
;
12
:
266
79
.
16.
Bosman
FT CF
,
Hruban
RH
,
Theise
ND
.
WHO classification of tumours of the digestive system. Lyon, France. The World Health Organization
. 4th ed.; 
2010
.
17.
Hruban
RH
,
Adsay
NV
,
Albores-Saavedra
J
,
Anver
MR
,
Biankin
AV
,
Boivin
GP
, et al
Pathology of genetically engineered mouse models of pancreatic exocrine cancer: consensus report and recommendations
.
Cancer Res
2006
;
66
:
95
106
.
18.
Hothorn
T
,
Zeileis
A
. 
Generalized maximally selected statistics
.
Biometrics
2008
;
64
:
1263
9
.
19.
Team
RC
.
R: A language and environment for statistical computing
.
R Foundation for Statistical Computing
,
Vienna
,
Austria
; 
2014
.
20.
Hothorn
T HK
,
van de Wiel
MA
,
Zeileis
A
. 
Implementing a class of permutation tests: the coin package
.
J Stat Soft
2008
;
28
1
23
.
21.
Harrell
FE
.
Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis
.
New York: Springer Science & Business Media
; 
2013
;
1
11
.
22.
Waddell
N
,
Pajic
M
,
Patch
AM
,
Chang
DK
,
Kassahn
KS
,
Bailey
P
, et al
Whole genomes redefine the mutational landscape of pancreatic cancer
.
Nature
2015
;
518
:
495
501
.
23.
Witkiewicz
AK
,
McMillan
EA
,
Balaji
U
,
Baek
G
,
Lin
WC
,
Mansour
J
, et al
Whole-exome sequencing of pancreatic cancer defines genetic diversity and therapeutic targets
.
Nat Commun
2015
;
6
:
6744
.
24.
Bailey
P
,
Chang
DK
,
Nones
K
,
Johns
AL
,
Patch
AM
,
Gingras
MC
, et al
Genomic analyses identify molecular subtypes of pancreatic cancer
.
Nature
2016
;
531
:
47
52
.
25.
Daemen
A
,
Peterson
D
,
Sahu
N
,
McCord
R
,
Du
X
,
Liu
B
, et al
Metabolite profiling stratifies pancreatic ductal adenocarcinomas into subtypes with distinct sensitivities to metabolic inhibitors
.
Proc Natl Acad Sci U S A
2015
;
112
:
E4410
7
.
26.
Jacobetz
MA
,
Chan
DS
,
Neesse
A
,
Bapiro
TE
,
Cook
N
,
Frese
KK
, et al
Hyaluronan impairs vascular function and drug delivery in a mouse model of pancreatic cancer
.
Gut
2013
;
62
:
112
20
.
27.
Lee
JJ
,
Perera
RM
,
Wang
H
,
Wu
DC
,
Liu
XS
,
Han
S
, et al
Stromal response to Hedgehog signaling restrains pancreatic cancer progression
.
Proc Natl Acad Sci U S A
2014
;
111
:
E3091
100
.
28.
Torphy
RJ
,
Volmar
KE
,
Naim
R
,
Cader
SR
,
Johansen
JS
,
Hollingsworth
MA
, et al
Quantification of tumor stroma as a biomarker in pancreatic adenocarcinoma
.
J Clin Oncol
2015
;
33
(
suppl; abstr 4021
).
29.
Koay
EJ
,
Truty
MJ
,
Cristini
V
,
Thomas
RM
,
Chen
R
,
Chatterjee
D
, et al
Transport properties of pancreatic cancer describe gemcitabine delivery and response
.
J Clin Invest
2014
;
124
:
1525
36
.
30.
D'Onofrio
M
,
Gallotti
A
,
Mantovani
W
,
Crosara
S
,
Manfrin
E
,
Falconi
M
, et al
Perfusion CT can predict tumoral grading of pancreatic adenocarcinoma
.
Eur J Radiol
2013
;
82
:
227
33
.
31.
Epelbaum
R
,
Frenkel
A
,
Haddad
R
,
Sikorski
N
,
Strauss
LG
,
Israel
O
, et al
Tumor aggressiveness and patient outcome in cancer of the pancreas assessed by dynamic 18F-FDG PET/CT
.
J Nucl Med
2013
;
54
:
12
8
.
32.
Rosenkrantz
AB
,
Matza
BW
,
Sabach
A
,
Hajdu
CH
,
Hindman
N
. 
Pancreatic cancer: lack of association between apparent diffusion coefficient values and adverse pathological features
.
Clin Radiol
2013
;
68
:
e191
7
.
33.
Kurosawa
J
,
Tawada
K
,
Mikata
R
,
Ishihara
T
,
Tsuyuguchi
T
,
Saito
M
, et al
Prognostic relevance of apparent diffusion coefficient obtained by diffusion-weighted MRI in pancreatic cancer
.
J Magn Reson Imaging
2015
;
42
:
1532
7
.
34.
Klauss
M
,
Gaida
MM
,
Lemke
A
,
Grunberg
K
,
Simon
D
,
Wente
MN
, et al
Fibrosis and pancreatic lesions: counterintuitive behavior of the diffusion imaging-derived structural diffusion coefficient d
.
Invest Radiol
2013
;
48
:
129
33
.
35.
Muraoka
N
,
Uematsu
H
,
Kimura
H
,
Imamura
Y
,
Fujiwara
Y
,
Murakami
M
, et al
Apparent diffusion coefficient in pancreatic cancer: characterization and histopathological correlations
.
J Magn Reson Imaging
2008
;
27
:
1302
8
.
36.
Wang
Y
,
Chen
ZE
,
Yaghmai
V
,
Nikolaidis
P
,
McCarthy
RJ
,
Merrick
L
, et al
Diffusion-weighted MR imaging in pancreatic endocrine tumors correlated with histopathologic characteristics
.
J Magn Reson Imaging
2011
;
33
:
1071
9
.
37.
Hansmann
J
,
Hernando
D
,
Reeder
SB
. 
Fat confounds the observed apparent diffusion coefficient in patients with hepatic steatosis
.
Magn Reson Med
2013
;
69
:
545
52
.
38.
Le Bihan
D
,
Breton
E
,
Lallemand
D
,
Aubin
ML
,
Vignaud
J
,
Laval-Jeantet
M
. 
Separation of diffusion and perfusion in intravoxel incoherent motion MR imaging
.
Radiology
1988
;
168
:
497
505
.
39.
Reeder
SB
,
Cruite
I
,
Hamilton
G
,
Sirlin
CB
. 
Quantitative assessment of liver fat with magnetic resonance imaging and spectroscopy
.
J Magn Reson Imaging
2011
;
34
:
729
49
.