Assessment of tissue perfusion of pancreatic cancer as potential imaging biomarker by means of Intravoxel incoherent motion MRI and CT perfusion: correlation with histological microvessel density as ground truth

Background/objectives The aim of this study was to compare intravoxel incoherent motion (IVIM) diffusion weighted (DW) MRI and CT perfusion to assess tumor perfusion of pancreatic ductal adenocarcinoma (PDAC). Methods In this prospective study, DW-MRI and CT perfusion were conducted in nineteen patients with PDAC on the day before surgery. IVIM analysis of DW-MRI was performed and the parameters perfusion fraction f, pseudodiffusion coefficient D*, and diffusion coefficient D were extracted for tumors, upstream, and downstream parenchyma. With a deconvolution-based analysis, the CT perfusion parameters blood flow (BF) and blood volume (BV) were estimated for tumors, upstream, and downstream parenchyma. In ten patients, intratumoral microvessel density (MVDtumor) and microvessel area (MVAtumor) were analyzed microscopically in resection specimens. Correlation coefficients between IVIM parameters, CT perfusion parameters, and histological microvessel parameters in tumors were calculated. Receiver operating characteristic (ROC) analysis was performed for differentiation of tumors and upstream parenchyma. Results ftumor significantly positively correlated with BFtumor (r = 0.668, p = 0.002) and BVtumor (r = 0.672, p = 0.002). There were significant positive correlations between ftumor and MVDtumor/ MVAtumor (r ≥ 0.770, p ≤ 0.009) as well as between BFtumor and MVDtumor/ MVAtumor (r ≥ 0.697, p ≤ 0.025). Correlation coefficients between ftumor and MVDtumor/ MVAtumor were not significantly different from correlation coefficients between BFtumor and MVDtumor/ MVAtumor (p ≥ 0.400). Moreover, f, BF, BV, and permeability values (PEM) showed excellent performance in distinguishing tumors from upstream parenchyma (area under the ROC curve ≥0.874). Conclusions The study shows that IVIM derived ftumor and CT perfusion derived BFtumor similarly reflect vascularity of PDAC and seem to be comparably applicable for the evaluation of tumor perfusion for tumor characterization and as potential quantitative imaging biomarker. Trial registration DRKS, DRKS00022227, Registered 26 June 2020, retrospectively registered. https://www.drks.de/drks_web/navigate.do?navigationId=trial. HTML&TRIAL_ID=DRKS00022227. Supplementary Information The online version contains supplementary material available at 10.1186/s40644-021-00382-x.


Introduction
Pancreatic ductal adenocarcinoma (PDAC) is notorious for its exceptionally high mortality with an estimated global mortality rate of 98% [1]. It has been reported that the complex tumor biology with the distinct microenvironment of the cancer has important influence on the prognosis of PDAC [2][3][4]. One of the factors which are known to be associated with prognosis in malignomas is tumor vascularity quantified histologically by the intratumor microvessel density (i.e. number of microvessels per tumor area, MVD). In various studies on PDAC, high MVD was associated with higher risk for liver and lymph node metastases as well as short survival time [5][6][7][8][9][10][11][12].
Tissue perfusion is regarded as a non-invasive marker for tumor vascularity and has been evaluated as biomarker of drug action in early phase trials for the treatment of various solid tumors [13], including PDAC [14]. Moreover, tissue perfusion was reported useful for the differentiation of local recurrence from unspecific postoperative alterations after PDAC resection [15].
For measurement of tumor tissue perfusion, the intravoxel incoherent motion (IVIM) model for diffusion weighted (DW) magnetic resonance imaging (MRI) as well as computed tomography (CT) perfusion have gained increasing interest during the past years. The IVIM model enables to extract parameters of both pure extravascular molecular diffusion and perfusion from DW MRI data [16]. CT perfusion measures tissue perfusion through the analysis of temporal changes in attenuation during the first pass of a bolus of iodinated contrast material [13].
Many previous studies have reported direct correlations between CT perfusion-derived parameters and MVD in other solid tumors [17], including pancreatic neuroendocrine neoplasms (PNEN) [18]. However, studies which correlate the histologically derived MVD with IVIM-derived parameters, or correlate IVIMderived parameters and CT perfusion-derived parameters are still quite rare. In a recent study, Klauß et al. showed a good correlation between the IVIM-derived perfusion fraction f and the MVD in PDAC and PNEN [19]. A previous study failed to show a correlation between IVIM-derived parameters and CT perfusionderived parameters in hepatocellular carcinomas [20].
This raises the question if CT perfusion and IVIM MRI are comparably applicable for the assessment of tissue perfusion in solid tumors, or whether one method should be preferred over the other.
Therefore, the aim of our study was to compare IVIM DW MRI and CT perfusion to assess tumor perfusion of PDAC.

Study design, patient population and demographics
The study protocol of our prospective study was approved by our institutional review board and informed consent was obtained from the patients before the examination. Between August 2014 and July 2015, 23 patients (13 females and 10 males; mean age, 63.2 ± 7.9 years [range: 50-79 years]) with detection of a mass suspicious of PDAC in a previous CT and/or MRI and scheduled surgery with potentially curative intent were consecutively enrolled in our study. In all patients, indication for surgery had been made prior to and independently from the present study. The patients were scanned and evaluated prospectively, first with IVIM DW MRI, and second with CT perfusion, on the day before surgery. Exclusion criteria were: general contraindications for the application of iodinated contrast agents, MR-unsafe foreign bodies, previous treatment for pancreatic carcinoma, inability to reproduce the breathing technique (see below), and/or denial of consent.
There was no histologically confirmed diagnosis at the time of inclusion in our study.
Out of the 23 total patients, the final histopathological diagnosis was PDAC in 20 patients, PNEN in one patient, anaplastic carcinoma in one patient, and massforming chronic pancreatitis in one patient. The three patients with histopathological diagnosis other than PDAC were excluded from our study. In one patient with PDAC the tumor area was missed in CT perfusion, therefore this patient was also excluded. CT perfusion and MRI diffusion could be evaluated in all remaining 19 patients. In 16 of these 19 patients, the tumor was located in the pancreatic head while 3 patients had a tumor in the pancreatic body and/ or tail. Among these 19 patients, resection was not possible and/or indicated due to infiltration of the superior mesenteric artery (SMA) in 1 patient, due to peritoneal metastases in 2 patients, and due to hepatic metastases in 1 patient. In these 4 patients in whom tumor resection was not performed, histological diagnosis was established from intraoperative biopsies from the part where the tumor infiltrated the SMA, or from the peritoneal/ hepatic metastases, respectively. In all 15 other patients, histological diagnosis was established from the resected primary tumor.

Null hypothesis and alternative hypothesis
The null hypothesis of the study is that the correlation of the perfusion fraction f from IVIM DW MRI with MVD/ microvessel area (MVA) is not different from the correlation of blood flow (BF) from CT perfusion with MVD/ MVA. The alternative hypothesis of the study is that the correlation of f with MVD/ MVA is different from the correlation of BF with MVD/ MVA.

MR imaging and Post-processing
MR imaging was performed using a 1.5 T scanner (Magnetom Avanto, Siemens Medical Solutions) with a 6element body-phased array coil and a 24-channel spine array coil. The pancreatic MR imaging protocol consisted of anatomic imaging sequences and diffusion weighted imaging (DWI) with 9 b-values (0, 50, 100, 150, 200, 300, 400, 600, and 800 s/mm 2 , summarized in Table 1). This combination of b-values was chosen since The acquisition was separated into blocks (b 0 , b 50 ), (b 0 , b 100 ) … (b 0 , b 800 ). Each block was acquired in a single breath-hold in expiration (TA = 22 s) to avoid motion artifacts. No registration for correction of patient breathing-motion was applied.
The pancreatic MR imaging protocol consisted of 1) anatomic imaging sequences, and 2) diffusion weighted imaging with 9 b-values. An experienced radiologist directly involved in the study was always present during MR imaging. Abbreviations: n.a.: not applicable, fs: fat saturation, HASTE: Half-Fourier-Acquired Singleshot Turbo spin Echo, ss-EPI: Single-shot Echo-Planar Imaging, TE: echo time, TR: repetition time it had proven feasible for assessment of tissue perfusion in correlation to MVD in PDAC [19]. b-Values larger than 800 s/mm 2 were not chosen to minimize the kurtosis effect that becomes increasingly important at larger b-values [21]. MITK Diffusion software version 2017.07 (Medical Imaging Interaction Toolkit, DKFZ Heidelberg, www. MITK.org) was used for post-processing of DWI data [22]. Among several possible approaches described by Klaasen et al. [23], the following approach according to the IVIM model was chosen to calculate the perfusion fraction f, pseudodiffusion coefficient D*, and diffusion coefficient D, as previously described [19] (corresponding to Klaasen's model no. 3). The signal was averaged within a region of interest for each b-value. Then the equation was fitted to the data. Here, S b stands for the signal with diffusion weighting and S 0 for the signal without diffusion weighting. Measurements at b-values greater than 170 s/mm 2 were used in a first step to estimate f and D. D* was then calculated in a second step by using exhaustive search. Quantitative analysis of DWI was performed independently by two radiologists with at least 5 years of experience in abdominal imaging each (P.M. and F.F.), blinded to the other radiologist's analysis and other patient information. Free-hand volumes of interest (VOIs) were drawn encompassing the tumor on DW images. The exact anatomical outline of the tumor was determined with the help of conventional CT images and/ or conventional biliary-pancreatic MR images. Calcifications (as detected by CT) and cystic/ necrotic tumor areas without enhancement (as detected by contrast enhanced CT) were excluded from the VOIs. When possible, upstream and downstream pancreatic parenchyma also was segmented by Reader 1 (P.M.). The reported values of f, D, and D* were derived from the generated VOIs in all cases.

CT imaging and Post-processing
Immediately after MR imaging, the patients were examined with CT imaging.
All examinations were carried out with a 2 × 64-slice CT scanner (Somatom Definition Flash; Siemens Medical Solutions), using the hydro-CT-technique [24]. Patients were placed on the CT table in an oblique, 30°, right-sided down position. The acquisition protocol is summarized in Table 2.
Perfusion data were analyzed with a body-perfusion CT-tool (Body-PCT, Siemens Medical Solutions) at a multimodality workplace with the syngo.via imaging software version VB 30 (Siemens Medical Solutions).
The baseline definition for motion correction at any time step and for segmentation at time step zero was followed by the segmentation of an organ VOI and the definition of a circular region of interest (ROI) in the aorta for vascular identification. The mean tissue timeattenuation curve was derived automatically and based on these definitions and data the color-coded parameter maps were established and confirmed.
Two radiologists with at least 5 years of experience in interpreting abdominal images each (P.M. and F.F.), independently placed polygonal VOIs encompassing the tumor, blinded to the other radiologist's analysis and other patient information. For exact CT VOI placement, the radiologists had access to the same set of conventional CT/MR images as provided for MRI VOI placement. When possible, upstream and downstream pancreatic parenchyma was also segmented by Reader 1 (P.M.). Calcifications and cystic/ necrotic tumor areas without enhancement were excluded from the VOIs. To avoid a potential bias, the time interval between IVIM DWI analysis and CT perfusion analysis was at least 3 months for each radiologist.
Using a deconvolution model the software calculated the following parameters: BF blood flow ð Þ ml 100 mlÃ min PEM permeability ð Þ ml 100 mlÃ min: The dose-length-products (DLPs) were calculated from the volume CT dose index (CTDI vol ) -values and scan lengths: For calculation of the effective dose (D eff ), the DLPs were multiplied with the corresponding conversion factor for abdominal CT-examinations [25]: Histology and immunohistochemistry for the assessment of the MVD In resection specimens, the diagnosis of PDAC was established according to the criteria recommended by the World Health Organization (WHO) and pathological staining was provided using the Union internationale contre le cancer (UICC) criteria. Histopathological

B30f
The pancreatic CT imaging protocol consisted of 1) a standard 3-phasic CT acquisition which was only performed if the patient didn't have an in-house CT examination of the abdomen within previous 4 weeks, 2) a native acquisition for verification of the correct position of the examination volume for CT perfusion imaging (performed in every patient), and 3) CT perfusion imaging (performed in every patient). All CT acquisitions were performed with automated tube-current modulation. CT acquisitions from examination series grouped under point 1) and point 2) were performed with automated selection of tube voltage (reference tube voltage listed). An experienced radiologist directly involved in the study was always present during CT imaging. Abbreviations: n.a.: not applicable grading was based on combined assessment of growth pattern, mucin content, and mitoses. When heterogeneity (i.e. variation in the degree of differentiation) was seen, the highest grade was assigned. TNM (tumor, node, metastasis) stages according to the 8th Edition of the UICC Manual and Grading (G) of the 15 resected tumors were as follows: T1 in 1 patient, T2 in 10 patients, T3 in 4 patients, N0 in 3 patients, N1 in 4 patients, N2 in 8 patients, M0/x in 14 patients, M1 in 1 patient. The histopathological grading was G2 in 8 patients, and G3 in 7 patients. Pathological tumor size in these 15 patients ranged from 1.8 cm to 6.1 cm (mean value: 3.35 cm ± 1.19 cm).
Because of the known tumor heterogeneity [3], tissue selection for tissue-based analysis was performed as previously described. In 10 out of 15 patients who underwent resection for PDAC, representative whole tumor slides from formalin-fixed paraffin-embedded tissue were immunostained with a CD34-specific antibody (1: 25, M7165, Dako), as previously described [19]. In 4 patients, tissue slides of non-neoplastic pancreatic tissue were also immunostained with the CD34-specific antibody.
To generate digital slide images, tissue slides were scanned at 20x magnification using an Aperio slide scanner (Leica Biosystems Aperio). In 10 patients, a mean coherent tumor area of 45 mm 2 per tumor, and in 4 patients, representative non-neoplastic pancreatic tissue were then analyzed using the Aperio Microvessel Analysis software (Leica Biosystems Aperio) [26], as previously described [19]. Plausibility was confirmed by pathologists (blinded to clinical information). MVA was calculated as the proportion of the sum of all vessel areas to the total analyzed area, MVD was calculated as mean vessel count per mm 2 .

Statistical analysis
Statistical data analysis was performed using MedCalc version 19.2.1 (MedCalc Software Ltd., Ostend, Belgium). Spearman rank correlation coefficients between IVIM-derived parameters, CT perfusion-derived parameters, and MVD/ MVA in tumors were calculated. As proposed by Campbell and Swinscow [27], Spearman correlation was interpreted as very weak (0.00-0.19), weak (0.20-0.39), moderate (0.40-0.59), strong (0.60-0.79), or very strong (0.80-1.00). For comparing correlation coefficients, we used the test recommended by Meng et al., Steiger's Z [28]. For a two-tailed test, Zscores greater than 1.96 or smaller than − 1.96 are considered statistically significant. Regression analysis was applied between f and BF, between f and MVD, as well as between BF and MVD, using linear regression models. Mann-Whitney U test was used for comparison of independent continuous variables while Wilcoxon test was used for comparison of dependent continuous variables. Inter-reader reliability was assessed by using the Intraclass Correlation Coefficient (ICC) with 95% confidence intervals (CI) and applying a 2-way ICC with random raters' assumption reproducibility. As proposed by Song et al. [29], ICC values were interpreted as poor (0.00-0.20), fair (0.21-0.40), moderate (0.41-0.60), good (0.61-0.80), or excellent (0.81-1.00). Receiver operating characteristic (ROC) curves were employed to analyze the diagnostic performance of DWI IVIM and CT perfusion parameters in distinguishing tumors from upstream parenchyma. Due to the small sample size of patients with downstream parenchyma (n = 5), ROC curve analysis was not performed for distinguishing tumors from downstream parenchyma. The AUCs with 95% confidence intervals (CIs) were computed. Sensitivities and specificities of the ROC curves were calculated, and the optimal cut-off values were determined. The DeLong method [30] was used for comparison of areas under the curves (AUCs). As proposed by Mandrekar [31], AUC values were interpreted as acceptable (0.70-0.79), excellent (0.80-0.89), and outstanding (0.90-1.00), while an AUC of 0.5 suggests no discriminatory ability. Significance level was set at 0.05.

Comparative data analysis: DWI IVIM and CT perfusion parameters in tumors versus downstream and upstream parenchyma
f upstream values and D upstream values from Reader 1 were significantly higher than f tumor values and D tumor values (p ≤ 0.0010). f downstream values were non-significantly higher than f tumor values (p = 0.0625). D downstream did not differ significantly from D tumor (p = 0.6250) and D* values did not differ significantly between tumors, downstream, and upstream parenchyma (p ≥ 0.6250). BF, BV, and PEM values were significantly lower in tumors than in upstream parenchyma (p ≤ 0.0003) and non-significantly lower in tumors than in downstream parenchyma (p = 0.0625).
Receiver operating characteristic (ROC) curves (with 95% CI) of DWI IVIM and CT perfusion parameters from Reader 1 for distinguishing tumors from upstream parenchyma are presented in Supplementary Figure 2

Comparative data analysis: DWI IVIM versus CT perfusion versus histological parameters
Spearman rank correlation coefficients a) between f tumor and BF tumor /BV tumor , as well as b) between MVD tumor / MVA tumor and f tumor /BF tumor /BV tumor are listed in Table 3. Spearman rank correlation coefficients between all tumor DWI IVIM-parameters, CT perfusion parameters and histological microvessel parameters are summarized in Supplementary Table 1.
There were significant positive strong Spearman rank correlation coefficients between mean f tumor for both readers and mean BF tumor / BV tumor (ρ ≥ 0.668, p ≤ 0.0018) in all patients (n = 19, Table 3). Moreover, there were significant positive strong to very strong Spearman rank correlation coefficients between mean MVD tumor /  Table 3). There were no significant Spearman rank correlations between mean D tumor / D* tumor for both readers and mean BF tumor / BV tumor in all patients (│ρ│ ≤ 0.392, p ≥ 0.0972) or between MVD tumor / MVA tumor and mean D tumor / D* tumor / PEM tumor in the 10 patients in whom histological microvessel analysis was performed (│ρ│ ≤ 0.527, p ≥ 0.1173). Also, the product f tumor x D* tumor , which supposedly reflects capillary blood flow [32], was not significantly correlated to BF tumor / BV tumor / MVD tumor / MVA tumor (│ρ│ ≤ 0.200, p ≥ 0.5617).
Correlation coefficients between mean MVD tumor / MVA tumor for both readers and mean f were not significantly different from correlation coefficients between mean MVD tumor / MVA tumor and mean BF tumor (p ≥ 0.4001, |Z-score| ≤ 0.840). Also, correlations between mean MVD tumor / MVA tumor and mean BF for both readers were not significantly different from correlations between mean MVD tumor / MVA tumor and mean BV tumor (p ≥ 0.5652, |Z-score| ≤ 0.575). D tumor , D* tumor , f tumor , BF tumor , and BV tumor values did not differ significantly between moderately differentiated (G2) and poorly differentiated tumors (G3, p ≥ 0.4495).
MVA and MVA values were higher in non-neoplastic parenchyma than in tumors for all 4 patients in whom

Discussion
The presented study evaluates tumor tissue perfusion as possible imaging biomarker in PDAC. In PDAC patients, tumor tissue perfusion was quantified by means of IVIM DW MRI as well as CT perfusion, and correlated with histologically determined MVD tumor / MVA tumor .
We used the deconvolution model for CT perfusion analysis since it can tolerate greater image noise than compartment models and was reported to be well suited for measuring lower levels of perfusion [33], as expected in PDAC. The deconvolution-based BF tumor values obtained in our current study were similar to previously reported deconvolution-based BF tumor values in PDAC [34], but considerably higher than previously reported BF tumor values based on the maximum-slope approach in PDAC [34,35]. This difference between BF tumor values obtained from the different models might be attributable to differences in the two mathematical calculation methods [34]. Mean BV tumor and PEM tumor values were within the same range as those reported by Schneeweiß et al. [34]. Also, mean values of the DWI IVIM parameters, as well as the histological microvessel parameters were within the same range as in a previous study by Klauß et al. [19].
The results of the current study confirm significant positive rank correlations between the IVIM-derived perfusion fraction f tumor and MVD tumor / MVA tumor in PDAC. Similarly to a study by Klauß et al. there was no significant rank correlation between D* tumor and MVD tumor or MVA tumor , although D* is supposed to be a flowrelated parameter [19]. A possible explanation for this observation could be linked to a study by Lemke et al. where the estimate of the pseudo-diffusion coefficient D* was found to be much less stable than the estimate of the perfusion fraction f [36]. As expected, the diffusion coefficient D tumor did not significantly correlate Fig. 2 Image examples of a patient with comparatively low tumor vascularity. 64 years old female patient with PDAC of the pancreatic corpus/ tail. a-c) Axial MR images with the patient in supine position. a Axial T2-HASTE MR image shows upstream dilatation of the main pancreatic duct and concomitant parenchymal atrophy (green arrows) due to an obstructing tumor in the pancreatic corpus/ tail (red arrows). b Diffusion-weighted MR image (b = 300 s/mm 2 ) with a VOI encompassing the tumor (VOI tumor). c Diffusion-weighted MR imaging with overlaying color-coded f-map. Mean calculated f tumor -value for both readers was low (8.9%). d-f Axial CT images with the patient in an oblique, 30°, right-sided down position. d Temporal maximum intensity projection (MIP) CT images of the perfusion sequence with VOIs encompassing the pancreas (VOI pancreas) and the tumor (VOI tumor). e-f Temporal MIP CT images with color-coded parameter maps for blood flow (BF tumor , e) and blood volume (BV tumor , f) derived from perfusion-sequence. Mean BF tumor and BV tumor were low (30.9 ml/100 ml/min and 1.8 ml/100 ml). g) Representative cutout of corresponding immunostained tissue slide (CD34) after semi-automated segmentation of microvessels shows low MVD tumor . MVD tumor in total analysis area was 34.1 /mm 2 with histological microvessel parameters in our present study. D represents perfusion-free diffusion [16] and was shown to reflect changes in the microarchitecture of the PDAC stroma [3].
The current study demonstrates that CT perfusion is a useful tool to evaluate tumor vascularity in PDAC. In the present study, BF tumor and BV tumor from CT perfusion similarly reflected MVD tumor in PDAC. However, in a study on PNEN, BF tumor significantly correlated with MVD tumor whereas BV tumor did not [18], which is in contrast to a study on colorectal cancer [37]. These discrepancies might be caused by differences in the microstructure of these tumor entities.
Rank correlations between microvessel parameters and PEM tumor from CT perfusion were not significant in the current study. This is logical since PEM tumor reflects features of microvessels like the microanatomy of the basement membrane rather than microvessel count [38]. Likewise, a study on PNEN failed to detect a significant correlation between PEM and MVD [18].
Probably the most important finding of the current study is that the applicability of DW IVIM MRI seems to be comparable to the applicability of CT perfusion for the assessment of tumor tissue perfusion. This finding is relevant for oncological imaging since tumor tissue perfusion can be used as imaging biomarker for treatment assessment [13]. This finding might also explain the excellent performance of both the perfusion fraction f from IVIM and CT perfusion parameters in distinguishing the notoriously hypovascular PDAC [39] from nonneoplastic pancreatic parenchyma in the present and previous studies [40,41]. Valuable statements on the applicability of different radiological methods for the assessment of tumor vascularity can be accomplished only by direct comparison of methods and validation with histological microvessel parameters as ground truth. Similar studies should be performed in other solid tumors to further evaluate the interrelations of DW IVIM MRI, CT perfusion, and histological tumor microvascularity.
There are some limitations to our study. First, our patient sample is relatively small, although the time spent and effort per patient were comparatively high. Second, the complete series of tissue slides, to overcome the tumor heterogeneity, for histological and microvessel analysis were only available in 10 out of the 15 patients, Fig. 3 Image examples of a patient with comparatively high tumor vascularity. 58 years old male patient with PDAC of the pancreatic head. a-c Axial MR images with the patient in supine position. a Axial T2-HASTE MR image shows upstream dilatation of the main pancreatic duct (green arrow) due to an obstructing tumor in the pancreatic head (red arrow). Stent in the distal common bile duct which is not dilated (pink arrowhead). b Diffusion-weighted MR image (b = 300 s/mm 2 ) caudad to the T2-HASTE image with a VOI encompassing the tumor (VOI tumor). c Diffusion-weighted MR imaging with overlaying color-coded f-map. Mean calculated f tumor -value for both readers was high (22.1%). d-f Axial CT images with the patient in an oblique, 30°, right-sided down position. D) Temporal maximum intensity projection (MIP) CT images of the perfusion sequence with VOIs encompassing the pancreas (VOI pancreas) and the tumor (VOI tumor). e-f Temporal MIP CT images with colorcoded parameter maps for blood flow (BF tumor , e) and blood volume (BV tumor , f) derived from perfusion-sequence. Mean BF tumor and BV tumor were comparatively high (77.3 ml/100 ml/min and 8.2 ml/100 ml). g Representative cutout of corresponding immunostained tissue slide (CD34) after semi-automated segmentation of microvessels shows high MVD tumor . MVD tumor in total analysis area was 61.8/mm 2 who underwent surgery with curative intent for PDAC. Strikingly, this number of patients was still sufficient to prove significant correlations between microvessel parameters and f tumor as well as BF tumor and BV tumor . Third, the complete series of tissue slides was inspected in these 10 patients and representative tissue samples for microvessel analysis were chosen. However, these tissue samples weren't completely congruent to the VOIs of the IVIM and CT perfusion analysis.

Conclusion
In conclusion, this study shows that f tumor and BF tumor / BV tumor similarly reflect microvasculature in PDAC and seem to be comparably applicable for the evaluation of tumor tissue perfusion for tumor characterization.