Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Pancreatic Intraepithelial Neoplasia Revealed by Diffusion-Tensor MRI

Pancreatic Intraepithelial Neoplasia Revealed by Diffusion-Tensor MRI ORIGINAL ARTICLE Pancreatic Intraepithelial Neoplasia Revealed by Diffusion-Tensor MRI Carlos Bilreiro, MD, Francisca F. Fernandes, MSc, Rui V. Simões, PhD, Rafael Henriques, PhD, Cristina Chavarrías, PhD, Andrada Ianus, PhD, Mireia Castillo-Martin, MD, PhD, Tânia Carvalho, DVM, PhD, Celso Matos, MD, and Noam Shemesh, PhD Conclusions: DTI and T2* are useful for detecting and characterizing PanIN in Objectives: Detecting premalignant lesions for pancreatic ductal adenocarci- genetically engineered mice and in the human pancreas, especially with AD and FA. noma, mainly pancreatic intraepithelial neoplasia (PanIN), is critical for early di- These are encouraging findings for future clinical applications of pancreatic imaging. agnosis and for understanding PanIN biology. Based on PanIN's histology, we hy- Key Words: diffusion tensor imaging, carcinoma in situ, pancreatic neoplasms, pothesized that diffusion tensor imaging (DTI) and T2* could detect PanIN. pancreatic intraepithelial neoplasia, animals, genetically modified Materials and Methods: DTI was explored for the detection and characteriza- tion of PanIN in genetically engineered mice (KC, KPC). Following in vivo (Invest Radiol 2024;00: 00–00) DTI, ex vivo ultrahigh-field (16.4 T) MR microscopy using DTI, T2* was per- formed with histological validation. Sources of MR contrasts and histological ancreatic cancer, mainly represented by ductal adenocarcinoma features were investigated, including histological scoring for disease burden (le- P (PDAC), is the third leading cause of cancer-related death in the sion span) and severity (adjusted score). To test if findings in mice can be trans- 1 United States, with an estimated 11% survival rate at 5 years. As symp- lated to humans, human pancreas specimens were imaged. toms develop late in the course of the disease, most patients are diagnosed Results: DTI detected PanIN and pancreatic ductal adenocarcinoma in vivo (6 at advanced stages precluding pancreatic resection surgery, which re- KPC, 4 KC, 6 controls) with high discriminative ability: fractional anisotropy mains the only potentially curative therapeutic option. Moreover, when (FA) and radial diffusivity with area under the curve = 0.983 (95% confidence in- patients are diagnosed in the early disease stages and subjected to tumor 3,4 terval: 0.932–1.000); mean diffusivity and axial diffusivity (AD) with area under resection, the 5-year survival rate reaches 25%–50%. Therefore, the curve = 1 (95% confidence interval: 1.000–1.000). MR microscopy with his- identifying precursor lesions for PDAC, mainly composed of pancreatic tological correlation (20 KC/KPC; 5 controls) revealed that sources of MR con- intraepithelial neoplasia (PanIN), could provide opportunities for early trasts likely arise from microarchitectural signatures: high FA, AD in fibrotic 5,6 diagnosis and development of more effective therapies. However, areas surrounding lesions, high diffusivities within cysts, and high T2* within le- PanINs are not diagnosed by current imaging modalities, with a single sions' stroma. The strongest histological correlations for lesion span and adjusted study describing small cystic lesions in magnetic resonance cholangio- score were obtained with AD (R =0.708, P < 0.001; R = 0.789, P < 0.001, respec- 7 pancreatography in a subset of patients with these lesions. The absence tively). Ex vivo observations in 5 human pancreases matched our findings in of noninvasive diagnostic tools also prevents the investigation of mice, revealing substantial contrast between PanIN and normal pancreas. PanIN's biology and tumorigenesis in humans, which remains largely unknown and mostly driven by research using genetically engineered 8–10 mouse models (GEMMs) and histopathological analyses. There- fore, there is an urgent need for developing imaging methods for PanIN Received for publication September 12, 2024; and accepted for publication, after revi- diagnosis and characterization, which could enable early diagnosis be- sion, October 11, 2024. From the Radiology Department, Champalimaud Foundation, Lisbon, Portugal (C.B., fore PDAC is established. C.M.); Champalimaud Research, Champalimaud Foundation, Lisbon, Portugal Diffusion-weighted MRI (DWI) provides insights into tissue mi- (C.B., F.F.F., R.H., C.C., A.I., M.C.-M., T.C., C.M., N.S.); Nova Medical School, crostructure and is routinely used for diagnosis, staging, and evaluation Lisbon, Portugal (C.B.); i3S—Instituto de Investigação e Inovação em Saúde, of response to therapy in benign and malignant diseases, mostly based Universidade do Porto, Porto, Portugal (R.V.S.); and Pathology Department, 11–14 Champalimaud Foundation, Lisbon, Portugal (M.C.-M.). on apparent diffusion coefficients (ADCs). Given the potential of Conflicts of interest and sources of funding: N.S. serves on the Scientific Advisory DWI, diffusion tensor imaging (DTI) was developed and explored in Board of Bruker Biospin. No other potential conflicts of interest are reported from the clinical setting to provide a rotationally invariant representation of remaining authors. Champalimaud Foundation funded this work; R.V.S.'s work the diffusion process in biological tissues. In the pancreas, DWI has was supported by H2020-MSCA-IF-2018, ref: 844776; A.I.'s work received the support of a fellowship from “La Caixa” Foundation (ID 100010434) and from been used in the past for characterizing benign and malignant pro- the European Union's Horizon 2020 research and innovation programme under cesses, but studies exploring DTI for pancreatic disease characteriza- the Marie Skłodowska-Curie grant agreement no. 847648, fellowship code LCF/ tion remain lacking, including for the detection of PanIN and BQ/PI20/11760029. 15,16 early PDAC. Correspondence to: Noam Shemesh, PhD, Champalimaud Research, Champalimaud Foundation, Av. Brasilia 1400-038, Lisbon, Portugal. E-mail: noam.shemesh@ We hypothesized that high-resolution DTI, with its ability to neuro.fchampalimaud.org. characterize diffusion anisotropy and sensitivity toward microarchitecture, Data availability statement: The data generated in this study are available upon request from could detect and characterize microstructural changes expected in the corresponding author. Any human-derived data are completely anonymized. PanIN and early PDAC. To test this hypothesis, we imaged GEMM de- Supplemental digital contents are available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article veloping PanIN and PDAC in vivo and harnessed ex vivo MR micros- on the journal’s Web site (www.investigativeradiology.com). copy at ultrahigh field and direct histological validation to understand Copyright © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. This is an the sources of contrast. Similar techniques were previously successful open-access article distributed under the terms of the Creative Commons 17,18 in characterizing malignant lymph nodes with high resolution. Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download and share the work provided it is properly cited. The Here, we used 2 well-established mouse models of pancreatic cancer: work cannot be changed in any way or used commercially without permission KC, mostly displaying PanIN lesions, and KPC, with complete pene- from the journal. trance of PDAC. Finally, we tested our hypothesis in human ISSN: 1536-0210/24/0000–0000 pancreas samples. DOI: 10.1097/RLI.0000000000001142 Investigative Radiology � Volume 00, Number 00, Month 2024 www.investigativeradiology.com 1 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 are described in Table 2. Anesthesia was induced with 5% isoflurane MATERIALS AND METHODS (Vetflurane; Virbac, France) mixed with oxygen-enriched (28%) air. Ethics Statement Isoflurane dosage was reduced to 1.5%–2.5% during animal preparation and for the duration of MRI scanning. Breathing rate and rectal temper- Animals were handled in agreement with European FELASA ature were monitored and recorded throughout scans using a pillow pres- guidelines. Animal procedures were conducted according to European Di- sure sensor and an optic fiber temperature probe (SA Instruments Inc, rective 2010/63 and preapproved by institutional and national authorities. Stony Brook, USA). A warm-water recirculating pad was used for body The study protocol involving human biological tissue was preapproved temperature control, maintaining rectal temperature between 35.0°C and by the institutional ethics committee. All participants provided signed 37.0°C. Ophthalmic gel (Vidisic Gel; Bausch+Lomb, Canada) was ap- informed consent before tissue collection, and personal identity was plied to the animals' eyes before each acquisition for preventing eye dry- kept private through pseudonymization in the institution's Biobank. ness. All acquisitions began with routine adjustments: center frequency, Genetically Engineered Mouse Models radiofrequency calibration, and automatic shimming. G12D G12D The KC (Pdx1-Cre, LSL-Kras /Ptf1a-Cre, LSL-Kras ) G12D LoxP In Vivo DTI Animals Selection and KPC (Ptf1a-Cre, LSL-Kras ,p53 ) mouse models were KPC mice presented early-onset disease with pancreatic abnor- used: KC mice develop PanIN and progress to PDAC; KPC mice de- malities in anatomical T2-weighted in vivo imaging and physical signs velop PanIN and PDAC with complete penetrance. Healthy mice of malaise. In contrast, KC mice presented pancreatic abnormalities (C57BL/6 J background) were used as controls. All animals' rearing later in their development. Considering these preliminary findings, we was undertaken in a specific pathogen-free facility in a temperature- decided to use young animals from the KPC and older animals from controlled room, with a 12-hour light/dark cycle with ad libitum access the KC transgenic lines for this part of the study: 6 KPC, 4 KC, and 6 to food and water. The animals' cages were enriched with plastic igloos controls (Supplemental Digital Content 1, Table with sample character- and shredded paper. Given the potential of GEMM for developing ad- istics, http://links.lww.com/RLI/A980), with an estimated statistical vanced pancreatic disease, the animals were continuously monitored power of 0.95 (α = 0.05; estimated effect size = 2.2). The predicted for signs of malaise, namely, weight loss/cachexia, fur thinning, and ap- number of KC mice was 6; however, due to unexpected deaths of 2 athetic or aggressive behavior. Whenever these signs were observed, the KC mice before pancreas extraction, these animals could not be used animals were humanely sacrificed. for histopathological analysis and were excluded from the study. Anatomical In Vivo Imaging In Vivo DTI In a first exploratory phase (Fig. 1A), KC (n = 9) and KPC (n = 6) mice were imaged monthly from 4 months old, to assess the de- KPC (n = 6), KC (n = 4), and control (n = 6) mice were imaged in velopment onset of pancreatic abnormalities. Animals were imaged un- a 9.4 T scanner (BioSpec; Bruker, Germany; 40-mm ID linear transmit- til abnormalities were observed, with a T2-weighted turbo spin-echo se- receive volume coil). Intraperitoneal hyoscine butylbromide (5 mg/kg; quence (Table 1), in 1 T and 9.4 T scanners (respectively, Icon and Buscopan; Boehringer Ingelheim, Spain) was administered 8 minutes BioSpec; Bruker, Germany; 40-mm ID linear transmit-receive volume before acquisition for bowel motion reduction. A diffusion-weighted coil) depending on scanner availability: 41 scans were performed in spin-echo echo-planar-imaging was used for imaging the animals' ab- 1 T; 15 scans were performed in 9.4 T scanners. Animal characteristics domen (Table 1). Datasets were analyzed using in-house developed FIGURE 1. Study design, diffusion-weighted multi-gradient-echo pulse sequence diagram, and preprocessing pipeline. A, Study design with major steps performed in chronological order. B, Diffusion-weighted multi-gradient-echo sequence used for the ex vivo MRI acquisition, which includes a small flip- angle slab-selective excitation followed by diffusion gradients and phase encoding, and a multiple-gradient-echo readout echo-train (RF, radiofrequency; SS, slice selection; PE, phase encoding; RO, readout; GDiff, diffusion gradients; δ, diffusion gradient duration; Δ, diffusion time; TE2 = TE3 = TE4). C, Data preprocessing pipeline, with a respective example image obtained in each step (MP-PCA = Marchenko-Pastur principal component analysis). Difference map is the resulting subtraction of pixel values between the raw image and the processed image after denoising and unringing, displaying the removal of image noise and ringing artifacts. 2 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. Investigative Radiology Volume 00, Number 00, Month 2024 DTI Detects Pancreatic Precancerous Lesions TABLE 1. MRI Acquisition Parameters Pulse Sequence T2-TSE SE-EPI DMGE Magnetic field strength 1 T; 9.4 T 9.4 T 16.4 T In-plane resolution (mm ) 0.150  0.150 (1 T) 0.125  0.125 - 0.125  0.125 (9.4 T) Slice thickness (mm) 0.7 (1 T) 0.25 - 0.3 (9.4 T) No. slices 20–36 36 - 3D resolution (mm) - - 0.08  0.08  0.08 TR (ms) 3825 (1 T) 2300–2600 125 1500–2700 (9.4 T) TE (ms) 60 (1 T) 15.6–16.5 9.2–10.2 20–22 (9.4 T) Signal averages 30 (1 T) 16 1 10–16 (9.4 T) Acceleration factor 8 - - Multishot segments - 4 - Partial Fourier reduction factor - 1.7 - EPI sampling bandwidth (kHz) - 395 - Fat suppression No Yes Yes No. DTI B values - 1 1 Bvalues (s/mm ) - 1000 775–1200 No. B = 0 s/mm -2 2 Diffusion directions - 10 10 Respiratory triggering Yes Yes - Scan duration 44–50 min (1 T) 31–33 min 5 h 54 min–16 h 33 min 12–21 min (9.4 T) T2-TSE, T2-weighted turbo spin echo; SE-EPI, spin-echo echo planar imaging; DMGE, diffusion-weighted multi-gradient-echo. code in MATLAB (MathWorks Inc, Natick, MA). Data preprocessing abnormalities were evident in T2-weighted images at 8.5 (interquartile included denoising based on Marchenko-Pastur principal component range [IQR] = 0.75) months of age for KC mice, these animals were 21,22 analysis (MP-PCA) and Gibbs unringing. Fractional anisotropy studied at this age (n = 12), whereas n = 8 KPC mice were studied at (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusiv- 4.25 (IQR = 0.25) months old, right after the first exploratory MRI ity (RD) maps were produced, fitting DTI to diffusion-weighted images scans, due to early aggressive disease. Preliminary findings were used using a weighted-least-squares solution. The images obtained were for calculating the number of animals in each group: pancreases har- analyzed by a gastrointestinal radiologist, who segmented the entire vested from 3 KC mice and 1 healthy control were imaged and FA maps pancreas in each slice, using MATLAB's built-in segmentation tools. were produced. According to a comparison of median FA measure- Measurements obtained from pancreatic segmentation were used for ments, effect size was estimated at 1.6, and a similar effect size was as- groupwise comparisons. After MRI, all animals had their pancreases sumed for KPC mice. With a statistical power of 0.95, each animal harvested and subjected to histopathological diagnosis, so disease status group was initially aimed at 10 animals. However, due to unpredictable (presence of PanIN, PDAC) was known for each animal. events when developing the colonies, 12 KC (available surplus in litter production) and 8 KPC mice were finally used (3 KPC mice suffered premature deaths and their pancreases could not be harvested on time). Pancreas Specimens' Preparation for MR Microscopy GEMM and n = 5 controls were sacrificed with cervical dislocation, Pancreases from additional KPC (n = 8), KC (n = 12), and and their pancreases dissected and removed through median laparot- control (n = 5) mice were prepared for MR microscopy. As pancreatic omy. Pancreatic specimens were subjected to fixation: immersion in TABLE 2. Sample Characteristics in Anatomical Longitudinal T2-Weighted Imaging of GEMM TABLE 3. Lesion Adjusted Score Calculation, Adapted From 25 26 Veite-Schmahl et al and Basturk et al Animal Model KC KPC Score Grade Distribution No. animals 9 6 Pancreatic abnormalities 8.5 (IQR = 0.75) 4.25 (IQR = 0.25) 1 Low-grade Focal onset (months old) 2 High-grade Multifocal Animal weight (g) 30.4 (IQR = 11.2) 22.9 (IQR = 4.1) 3PDAC Diffuse Animal sex 4/9 females 4/6 females The final adjusted score for each pancreatic specimen was determined using the Continuous variables expressed as medians and IQR. Pancreatic abnormalities following distribution-adjusted lesion grade = (low-grade  distribution) + (high- onset was assessed from 4 months old. grade  distribution) + (PDAC  distribution). © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. www.investigativeradiology.com 3 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 4% paraformaldehyde for 48 hours and then in phosphate-buffered sa- distribution) + (PDAC  distribution) with ascribed scores of 1, 2, line with 0.01% sodium azide for 24 hours. After fixation, pancreases and 3 for low-grade PanIN, high-grade PanIN, and PDAC, respectively, were immersed in Fluorinert within a 10-mm NMR tube for ex vivo and for focal, multifocal, and diffuse lesions, respectively. The same MRI, with special care for removing air bubbles and animal hair in histological slide was also used to calculate lesion span (ie, lesion bur- the sample. den for each specimen), through digital segmentation using QuPath v. 0.4.3 semimanual segmentation tools, by annotating area of the lesions 28,29 Ex Vivo MR Microscopy and area of normal pancreas. Pancreases were imaged in a 16.4 T MRI scanner (Ascend Aeon; Lesion span and adjusted score were correlated with MRI data, Bruker, Germany), using a diffusion-weighted multi-gradient-echo se- assessing if DTI and T2* measurements scale with disease extension quence, providing ultrahigh-resolution 3D images with both diffusion and severity. and T2* weighting (Fig. 1B, Table 1). Two 10-mm probes were used (Micro5, Cryoprobe; Bruker, Germany) depending on availability; sam- Imaging and Histology Alignment ple temperature was maintained at 37°C. Datasets were preprocessed as The alignment between MRI and histology was manually per- previously described (MP-PCA denoising, Gibbs unringing, 21,22 formed by a gastrointestinal radiologist. For this, the Volume Viewer Fig. 1C). After preprocessing, images were masked to remove tool from ImageJ (US National Institutes of Health) was used for low-signal areas. T2*, DTI (FA, MD, AD, RD), and ADC maps were reconstructing the 3D datasets in real time and searching for image then obtained from the preprocessed datasets. planes corresponding to histology slides using the powder-averaged dif- Histological Analysis fusion-weighted images. Specimens' structures (lymph nodes, large blood vessels, cystic lesions, and large masses) were used as landmarks Following MRI, mouse pancreas samples were serially sectioned (4-μm thickness, 0.5-mm spacing) and stained with hematoxylin and for searching for the most adequate image plane. When the manually re- eosin. Slides were digitally scanned in Ultra-Fast Scanner (Philips) or constructed image plane and slice were deemed the most approximate AxioScan Z1 (Zeiss) and examined by a veterinary pathologist. as possible to the corresponding histological slide, the reconstruction The classification of pancreatic lesions—acinar-to-ductal meta- coordinates were used in all DTI and T2* maps, and the corresponding plasia (ADM), low-grade PanIN, high-grade PanIN, and PDAC—was images were exported for a visual comparison between both histology based on specific criteria (Supplemental Digital Content 2, Table with and MRI. Image contrasts were therefore attributed to the correspond- classification, http://links.lww.com/RLI/A980), adapted from Veite- ing histological changes, as directly observed. 25 26 Schmahl et al and Basturk et al. Disease severity scoring was per- formed on the slide where the extent of the lesions was higher. A Human Pancreas distribution-adjusted lesion grade with equal contribution from each le- sion type was developed, ensuring that all lesions contribute equally to Pancreas specimens (n = 5) obtained from patients with PDAC, the final score, rendering a continuous range of scores, adapted from from the institution's Biobank (formalin-fixed leftovers after diagnostic pro- 27 28 30 Berman-Booty et al and Knoblaugh and Himmel. This system cal- cedures following resection surgery), were analyzed searching for PanIN. culates the score by summing the product of the scores for grade and Specimens were sliced and placed inside a 10-mm MR tube. Sample prep- distribution of each lesion, as shown in Table 3: distribution-adjusted le- aration, imaging acquisition, and processing methods described above for sion grade = (low-grade PanIN  distribution) + (high-grade PanIN- MR microscopy were used. Histological slides were sectionized, stained, FIGURE 2. Pancreatic abnormalities with cystic changes are seen in vivo in KPC (4.75 months old) and KC (8.5 months old) mice, using conventional TSE- T2-weighted images, with corresponding histological diagnosis, including acinar-to-ductal metaplasia, PanIN, and PDAC lesions. The healthy control (2 months old) pancreas is homogeneous without focal or diffuse changes. C, A necrotic/cystic area inside a large tumor (arrows) is directly correlated with histology. All 3 images were acquired in a 9.4 T scanner (BioSpec). 4 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. Investigative Radiology Volume 00, Number 00, Month 2024 DTI Detects Pancreatic Precancerous Lesions and digitized, as described above, and evaluated by a gastrointestinal pa- RESULTS thologist. Alignment was performed manually by a gastrointestinal radiol- ogist, as previously described, for a visual comparison. Longitudinal MRI Assessment of PanIN and PDAC Development Onset Exploratory anatomical in vivo T2-weighted images revealed Statistical Analysis variably enlarged pancreases with high signal intensity and cystic le- IBM SPSS statistics v.23 was used. Kolmogorov-Smirnov test was sions for GEMM, whereas the pancreas of control mice was homoge- used to assess the distribution of continuous variables, revealing nonnormal neous with low signal intensity (Fig. 2A). These findings were evident distributions throughout. Kruskal-Wallis test was used for groupwise com- at 8.5 months old for KC mice and at first scans (4.25 months old) for parisons, with post hoc pairwise comparisons, and Mann-Whitney U test KPC mice (Table 2), as confirmed with histopathology (Fig. 2B). for comparisons between 2 groups. Bonferroni correction was applied when comparing multiple test metrics, namely, different DTI, T2*, and DTI Noninvasively Detects PanIN and PDAC In Vivo ADC maps, by multiplying the P values by the number of metrics. Re- ceiver operating characteristics analysis was performed for assessing the Both KC and KPC mice presented pancreases with high FA, discriminative value of DTI. Pearson correlation coefficient was used for MD, AD, and RD; large PDAC; and small cystic lesions, contrasting vi- evaluating MRI measurements and lesion span/adjusted score. All tests sually with control mice which presented low values in the pancreas in were bilateral; P values <0.05 were considered significant. all parameters (Fig. 3A). FIGURE 3. DWIs (averaged b = 0 and b = 1000 s/mm images) and DTI maps (FA, MD, AD, RD) for representative KC, KPC, and healthy control mice. A, The maps produced with DTI showcase the differences between healthy controls, KC, and KPC mice. The KC presented here, with diffuse PanIN infiltration in the pancreatic parenchyma and no PDAC, has visually increased values in DTI derived maps, especially FA and AD. The KPC mouse is bearing a large PDAC, also with increased values in the same maps. In contrast, the healthy control mouse has a homogeneous pancreas with intermediate-low values in all maps. B, Quantitatively, the differences between groups of animals, including 10 GEMM (6 KPC and 4 KC) and 6 controls, provide a good discriminative ability of DTI-derived metrics. FA presents the best performance when using median values (AUC = 0.983); AD and MD present the best performance when using 95th percentile values (AUC = 1). MD, AD, RD expressed in mm /s. © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. www.investigativeradiology.com 5 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 GEMM and controls (Fig. 5; Supplemental Digital Content 4, Figure TABLE 4. In Vivo DTI With KC and KPC Mice—Presence of PDAC with all group comparisons, http://links.lww.com/RLI/A981). When comparing KPC and controls, median values were significantly different, PDAC Present Absent P with KPC presenting higher values for MD (+10.6%, P = 0.004), AD (+22%, P = 0.001), and RD (+7.7%, P = 0.023). When comparing KC No. animals 7 3 - and controls, only median AD was different (+14%, P = 0.046). The com- Models KC: 1/4 KC: 3/4 0.033* parison of highest values in each specimen (95th percentile) enhanced the KPC: 6/6 KPC: 0 observed differences between GEMM and controls. For KPC versus con- FA (median) 0.25 0.28 0.833 trols, except for FA (P = 0.128), KPC presented higher MD (+41.2%, FA (95th percentile) 0.45 0.48 0.383 P = 0.006), AD (+45.4%, P = 0.0018), RD (+38.9%, P = 0.004), T2* MD (median) 1.6E-3 1.5E-3 1.000 (+69.6%, P = 0.017), and ADC (+43.2%, P = 0.005). For KC versus con- MD (95th percentile) 2.1E-3 2.0E-3 0.517 trols, KC presented higher FA (+186.7%, P = 0.003), MD (+38.2%, AD (median) 2.1E-3 1.9E-3 0.517 P = 0.006), AD (+47.5%, P = 0.002), RD (+35.7%, P = 0.015), T2* AD (95th percentile) 2.5E-3 2.5E-3 0.833 (+58.6%, P = 0.010), and ADC (+40.8%, P = 0.006). When comparing RD (median) 1.3E-3 1.3E-3 1.000 KC and KPC, no differences were observed for median or 95th percentile RD (95th percentile) 1.9E-3 1.8E-3 0.517 values; however, the lowest values (5th percentile) were lower in KC for MD (−15.8%, P = 0.003), AD (−14.4%, P = 0.007), and RD (−15.2%, Continuous variables were expressed as medians. P = 0.005). Variable median ADC for different diffusion directions in each Fisher exact test was used for comparing prevalence of PDAC in GEMM animal further demonstrates diffusion anisotropy in the pancreas (Supple- models. mental Digital Content 5, Figure with median ADC for all diffusion Mann-Whitney U test was used for comparing GEMM with and without directions, http://links.lww.com/RLI/A982). PDAC for each DTI metric. Histologically, ADM was present in all GEMM mice, most *Statistically significant results. markedly in KC mice, as shown in Supplemental Digital Content 6 (Figure, representative segmented samples, http://links.lww.com/RLI/ A983) and Table 5 (all samples). PDAC was mainly observed in KPC Median FA obtained after pancreatic segmentation was signifi- mice: 33.33% of KC and 87.5% of KPC mice. Both GEMM had PanIN cantly higher for both transgenic mice (n = 6 KPC, n = 4 KC) compared lesions of grades 1 to 3, more marked in KC mice. Most mice had with controls (n = 6), with differences increasing and becoming signifi- ≥70% of their pancreas affected: 75% of KC mice and 75% of KPC cant also for MD, AD, and RD, when assessing the higher values ob- mice. The median lesion span in KC mice was 80.88% (IQR: 42.75) served in each group (ie, 95th percentile) (Fig. 3B). The discriminative and in KPC mice was 91.45% (IQR: 36.62). Median adjusted score performance of median DTI measurements obtained from the entire pan- was 17 (IQR: 9) in KC mice and 26 (IQR: 1.75) in KPC mice. creas in each animal for differentiating GEMM from controls was high, ranging from 0.767 (95% confidence interval [CI]: 0.527–1.000) area un- der the curve (AUC) for RD, up to 0.983 (95% CI: 0.932–1.000) AUC for DTI and T2* Correlate With Disease Burden FA. The discriminative performance increased once more when assessing and Severity the parameters' 95th percentile, with AUC ranging from 0.983 (FA and As seen in Figure 5, Supplemental Digital Content 7 (http:// RD, 95% CI: 0.932–1.000) to 1.000 (MD and AD, 95% CI: links.lww.com/RLI/A984), 8 (http://links.lww.com/RLI/A985) (Fig- 1.000–1.000) (Fig. 3B). When comparing KC and KPC mice, the DTI ures, lesion span and adjusted score correlation plots), and 9 (Table, le- metrics were similar between both groups of animals. Also, DTI metrics sion span and adjusted score correlation coefficients, http://links.lww. could not distinguish between GEMM with (n = 7) and without PDAC com/RLI/A980), DTI/T2* metrics correlate both with disease burden (n = 3) (Table 4). (lesion span) and severity (adjusted score), when using median and 95th percentile values. The highest correlation coefficients both for le- MR Microscopy With DTI and T2* Characterizes Ex Vivo sion span and for adjusted score were obtained with AD (median AD, R = 0.789; 95th percentile AD, R = 0.729; respectively). Other than Pancreatic Abnormalities FA and T2*, 5th percentile values of MRI metrics were not correlated To better understand the contrasts observed, we performed de- with histological scores. tailed MR microscopy and histology in pancreases harvested from KC mice (n = 12), KPC mice (n = 8), and healthy control mice (n = 5) PanIN in the Human Pancreas Revealed by Ex Vivo DTI (Supplemental Digital Content 3, Table with sample characteristics, http:// links.lww.com/RLI/A980). Human pancreas samples (Supplemental Digital Content 10, Ta- After alignment between MR microscopy and histology slides, ble with patient characteristics, http://links.lww.com/RLI/A980) pre- enabling a direct comparison of MRI and histology, stark contrasts be- sented relevant histological differences when compared with the tween DTI and T2* parameters in normal and abnormal tissue were ob- GEMM counterparts: scattered foci of ADM, higher proportion of nor- served (Fig. 4). High FA values are observed in areas of ADM, where mal pancreatic parenchyma, and absence of PDAC (Fig. 6, Supplemen- PanIN lesions of variable degrees are found surrounded by fibrosis, tal Digital Content 11, http://links.lww.com/RLI/A986). In 3 samples, as well as at the periphery of PDAC lesions. In contrast, areas of normal parenchymal atrophy with fatty replacement was observed (Figs. 6B parenchyma demonstrate low FAvalues, allowing a clear distinction be- and C, Supplemental Digital Content 11B, http://links.lww.com/RLI/ tween different types of tissue. The MD, AD, RD, and ADC maps reveal A986). After alignment between MR microscopy and histology, we ob- high values in cystic and necrotic areas, most evident in AD maps, and served clear contrasts between areas of ADM, PanIN, and surrounding intermediate/high AD values in fibrotic areas of ADM, PanIN, and normal pancreas. High FA, AD, and T2* values were observed in areas PDAC. T2* maps present high values in the stroma of both PanIN of ADM and surrounding PanIN lesions; contrast between lesions and and PDAC lesions. All the control pancreases show homogeneous surrounding parenchyma was evident in all maps. One pancreas sample FA, diffusivities, T2*, and ADC maps with lower values throughout with diffuse interstitial fibrosis, attributed to chronic pancreatitis, the pancreatic parenchyma. showed mid-high FA, AD, and T2* values irregularly scattered through- Quantitatively, as observed for in vivo DTI, DTI/T2* maps ob- out the fibrotic tissue (Fig. 6C). Histologically, this diffuse interstitial fi- tained from pancreas specimens allowed us to distinguish between brosis is haphazardly organized, whereas the periductal fibrosis forms a 6 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. Investigative Radiology Volume 00, Number 00, Month 2024 DTI Detects Pancreatic Precancerous Lesions FIGURE 4. Comparison between MR microscopy and histology in GEMM. A, The healthy mouse pancreas is homogeneous, with low anisotropy, diffusivity, and T2* values; coil used: Cryoprobe. B, KC pancreas with extensive ADM and PanIN (black arrows), as well as PDAC (white arrows). Conventional diffusion weighting shows heterogeneous restricted diffusion. There are high anisotropy values in the periphery of areas with ADM, PanIN, and PDAC. High axial diffusivity is observed in cystic/necrotic areas, and intermediate diffusivity and high T2* values are seen in the lesions' stroma. Coil used: Cryoprobe. C, KPC pancreas with extensive PDAC (white arrows) and small areas of ADM and PanIN (black arrows). Conventional diffusion weighting is not helpful in delineating the pancreatic lesions. There are high anisotropy values in the periphery of PDAC and PanIN lesions; high diffusivity values in cystic/necrotic areas and regions with inflammatory infiltrates; and high T2* values in the lesions' stroma. Coil used: Micro5. AD is expressed in mm /s; T2* is expressed in milliseconds (ms). sheath or cuff-like structure around ducts in areas of ADM, PanIN, Here, we leverage GEMM for DTI detection of PanIN, PDAC, and and PDAC. associated pancreatic changes, suggesting DTI/T2* as potential bio- markers for disease onset, staging, or monitoring. Our results also high- DISCUSSION light the important variability between pancreatic structural changes in these models, probably caused by variable proportions of tumoral This study provides evidence establishing a potential role for stroma, ductal structures, and inflammatory infiltrates, which should DTI and T2* in characterizing PanIN and PDAC. We report that ex vivo be accounted for in future studies. and in vivo MRI detects PanIN-associated changes and PDAC, identi- The DTI and T2* maps provided more striking visual contrasts fying potential mechanisms for the contrast and underscoring their than conventional diffusion-weighted images, the method currently translational relevance for imaging the human pancreas. used in clinical practice. As observed by comparing MRI with histol- The process of carcinogenesis in PDAC follows a progressive ogy, these contrasts appear to be derived from specific microstructural development of PanIN, where genetic mutations and morphological 10,31 characteristics of the lesions: areas of anisotropy with high FA values changes are acquired until PDAC is finally established. There is a corresponded to the peripheral zones of PanIN, ADM, and PDAC, recognized absence of noninvasive detection methods for PanIN, pre- 7,32 where sheath/cuff-like periductal fibrosis is observed; areas of high dif- cluding research into PanIN biology, especially in humans. Most fusivities corresponded mostly to cystic and necrotic regions; areas of current knowledge of PanIN was achieved with studies performed in 9,31 high T2* corresponded mostly to stromal components. Also, AD per- GEMM. Thus, the development of noninvasive detection methods formed better than other metrics, including the clinical standard ADC, can not only inform about PanIN biology and tumorigenesis, but also for lesion detection and correlated more closely with disease extent provide opportunities for early therapeutic interventions or close and severity. These results hint to differences in tissue microstructure follow-up of patients, enabling PDAC diagnosis in resectable stages. © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. www.investigativeradiology.com 7 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 FIGURE 5. Group comparisons, lesion span, and adjusted score for DTI (FA and AD are shown) and T2* metrics. For group comparisons, the differences between GEMM and controls are highlighted when using the highest (95th percentile) values, where AD provided the largest differences. For lesion span correlation, the use of 95th percentile values improved the results, where AD was the most correlated metric. For adjusted score, both medians and the 95th percentile values showed significant correlations with disease burden and severity, but the 95th percentile values did not show an improvement when compared with median values. Again, AD provided the best performance (R = 0.622, when using the median values). The group comparisons, lesion span, and adjusted score correlations for all MRI metrics can be found in Supplemental Digital Content 4, 7, and 8. as the origin of the observed MRI contrasts and highlight the value of sented with increased size and parenchymal replacement by abnormal microstructural characterization through DTI, as previously described tissue. The presence of fatty tissue is not observed in our DTI/T2* data, 13,33,34 for other applications. as the acquisitions were performed with fat suppression, and no fat- PanIN remains an elusive lesion in humans. Noncommunicating sensitive pulse sequences were performed. The characterization of fatty pancreatic microcysts in magnetic resonance cholangiopancreatogra- replacement is an interesting point for future studies, especially if per- phy were reported to predict the presence of PanIN in pancreatic spec- formed on the human pancreas. imens, with 61% accuracy. Our findings show that DTI/T2* might be DTI has been performed successfully for imaging the pancreas useful in this context, as the contrasts seen in human pancreas allowed a in clinical context. However, its clinical role remains undetermined, clear detection of PanIN. Regarding differentiation between degrees of both for PDAC detection/characterization and for characterizing other PanIN and PDAC, we found a clear correlation between disease burden malignant lesions or nonmalignant pancreatic processes. Technical dif- and severity and DTI/T2* metrics; however, the visual contrasts were ferences between MRI in basic research and clinical imaging are obvi- mostly similar, both in vivo and ex vivo. This may question the specific- ous, and pulse sequence adaptations with loss of resolution are expected ity of DTI/T2* for visually differentiating low-grade from high-grade due to hardware constraints in clinical scanners and time constraints PanIN and PDAC, but further data are needed to address this question. when imaging patients. For PanIN detection, information loss with The presence of parenchymal atrophy with fatty replacement has lower resolutions in 1.5 T and 3 T scanners might be overcome with been reported previously, as a finding associated with PanIN and early the use of pulse sequences sensitive to microscopic anisotropy. Future 7,35,36 PDAC. Our findings in human pancreas specimens corroborate studies should investigate if DTI can be used in clinical context for these reports, with 3 out of 5 samples presenting fatty replacement. This PanIN detection and characterization, addressing the question of speci- was not observed on GEMM, however, as the animals' pancreases pre- ficity for high- and low-grade PanIN and PDAC, as this issue is most TABLE 5. Histological Analysis of KC, KPC, and Control Mice Used for MR Microscopy Animal Model Controls KC KPC ADM 0% 100% (12/12) 100% (8/8) Low-grade PanIN 0% 100% (12/12) 100% (8/8) High-grade PanIN 0% 100% (12/12) 100% (8/8) PDAC 0% 33.33% (4/12) 87.5% (11/12) Lesion span (%) 0% 80.88 (IQR = 42.75) 91.45 (IQR = 36.62) Adjusted score 0 17 (IQR = 9) 26 (IQR = 1.75) Low-grade PanIN includes PanIN 1 and PanIN 2. Continuous variables expressed as medians and IQR. 8 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. Investigative Radiology Volume 00, Number 00, Month 2024 DTI Detects Pancreatic Precancerous Lesions FIGURE 6. The contrasts observed with MR microscopy DTI in the human pancreas are similar to the ones seen in GEMM (2 remaining specimens are presented in Supplemental Digital Content 11). A, High anisotropy, high T2*, and low AD are seen in the periphery of a high-grade PanIN lesion, contrasting well with the surrounding normal pancreatic parenchyma (arrows). B, As observed previously in the KC/KPC mice, areas with ADM and low- grade PanIN have high anisotropy and AD values at their periphery, with high T2* values in the stroma (arrows). Large areas of fatty tissue are seen in the specimen. C, This specimen has extensive haphazardously deposited fibrotic deposition and fatty infiltration, consistent with chronic pancreatitis. There are low-grade PanIN with high anisotropy, intermediate AD, and T2* values (arrows), but the fibrosis has heterogeneous anisotropy, higher AD, and T2* values. Some areas of this specimen have irregular T2* values, probably due to fixation artifacts. AD is expressed in mm /s; T2* is expressed in milliseconds (ms). relevant for implementing therapeutic interventions. Eventual combina- ORCID ID tions of DTI with other diagnostic tools could be used for increasing spec- Noam Shemesh https://orcid.org/0000-0001-6681-5876 38,39 ificity, such as liquid biopsy and artificial intelligence. Also, it would be interesting to investigate if DTI could be used in conjunction with 3D REFERENCES histological mapping, for detecting specific genomic profiles of PanIN 1. Siegel RL, Miller KD, Fuchs HE, et al. Cancer statistics, 2022. CA Cancer J Clin. and PDAC. Multiparametric MRI could also be studied for characterizing 2022;72:7–33. premalignant lesions, harnessing information from different pulse se- 2. Ducreux M, Cuhna AS, Caramella C, et al. Cancer of the pancreas: ESMO clinical quences and including diffusion-based techniques like intravoxel incoher- practice guidelines for diagnosis, treatment and follow-up. Ann Oncol.2015;26 40–42 (Suppl 5):v56–v68. ent motion, which have already demonstrated feasibility. 3. Campbell F, Smith RA, Whelan P, et al. Classification of R1 resections for pancre- We acknowledge some limitations of this research work. First, atic cancer: the prognostic relevance of tumour involvement within 1 mm of a re- GEMM do not fully recapitulate morphological and biological features section margin. Histopathology. 2009;55:277–283. of human PanIN and PDAC, as PanIN lesions in the mouse are mostly 4. Springfeld C, Ferrone CR, Katz MHG, et al. Neoadjuvant therapy for pancreatic present in the context of extensive ADM. Nevertheless, the observed cancer. Nat Rev Clin Oncol. 2023;20:318–337. image contrasts are useful for in vivo animal imaging, and large similar- 5. Ryan DP, Hong TS, Bardeesy N. Pancreatic adenocarcinoma. NEngl J Med. 2014;371:1039–1049. ities with human pancreas samples were observed, suggesting relevance 6. Overbeek KA, Goggins MG, Dbouk M, et al. Timeline of development of pancre- for human imaging. Second, the manual alignment between MRI and atic cancer and implications for successful early detection in high-risk individuals. histology was suboptimal and could be considered a potential source Gastroenterology. 2022;162:772–785.e4. of errors, as automatic registration was not viable due to the expected 7. Vullierme M-P, Menassa L, Couvelard A, et al. Non-branched microcysts of the distortion and configurational changes during sample processing for pancreas on MR imaging of patients with pancreatic tumors who had pancreatec- histology. Third, the in vivo application developed here demonstrates tomy may predict the presence of pancreatic intraepithelial neoplasia (PanIN): a the feasibility of the technique and its usefulness for differentiating preliminary study. Eur Radiol. 2019;29:5731–5741. GEMM from healthy controls but did not evaluate its role in determin- 8. Braxton AM, Kiemen AL, Grahn MP, et al. 3D genomic mapping reveals multifocality of human pancreatic precancers. Nature. 2024. Published online ing different degrees of disease progression in vivo. Also, KC and KPC May 1, 2024:;629:679–687. mice were observed in different age groups due to their different rates of 9. Alonso-Curbelo D, Ho YJ, Burdziak C, et al. A gene-environment-induced epige- disease progression. Future studies evaluating animals of the same netic program initiates tumorigenesis. Nature. 2021;590:642–648. transgenic line at different stages of disease progression should eluci- 10. Storz P, Crawford HC. Carcinogenesis of pancreatic ductal adenocarcinoma. Gas- date if DTI is useful for determining the stage of disease in similar troenterology. 2020;158:2072–2081. age groups. Finally, although similarities of imaging contrasts were ob- 11. Nilsson M, Englund E, Szczepankiewicz F, et al. Imaging brain tumour micro- served when imaging human pancreas samples, it remains to be deter- structure. Neuroimage.2018;182:232–250. mined if DTI can be successfully adapted to clinical scanners for imag- 12. Koh DM, Collins DJ. Diffusion-weighted MRI in the body: applications and chal- lenges in oncology. AJR Am J Roentgenol. 2007;188:1622–1635. ing patients and maintain sensitivity toward these lesions. Our work 13. Kiselev VG, Körzdörfer G, Gall P. Toward quantification: microstructure and provides exciting data in this regard and is encouraging for the develop- magnetic resonance fingerprinting. Invest Radiol.2021;56:1–9. ment of clinical DTI applications for pancreatic imaging. 14. Fukukura Y, Kanki A. Quantitative magnetic resonance imaging for the pancreas. In conclusion, PanIN lesions were detected with DTI and T2* Invest Radiol.2024;59:69–77. contrast in GEMM and human pancreas, suggesting high relevance as 15. Lanzman RS, Wittsack HJ. Diffusion tensor imaging in abdominal organs. noninvasive biomarkers. This augurs well for characterizing PanIN pro- NMR Biomed.2017;30. gression and biology and associated pancreatic changes in both mice 16. Nissan N, Golan T, Furman-Haran E, et al. Diffusion tensor magnetic resonance and humans. imaging of the pancreas. PloS One. 2014;9:e115783. © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. www.investigativeradiology.com 9 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 17. Santiago I, Santinha J, Ianus A, et al. Susceptibility perturbation MRI maps tumor 30. Andea A, Sarkar F, Adsay VN. Clinicopathological correlates of pancreatic intra- infiltration into mesorectal lymph nodes. Cancer Res. 2019;79:2435–2444. epithelial neoplasia: a comparative analysis of 82 cases with and 152 cases without pancreatic ductal adenocarcinoma. Mod Pathol. 2003;16:996–1006. 18. Ianuş A, Santiago I, Galzerano A, et al. Higher-order diffusion MRI characteriza- tion of mesorectal lymph nodes in rectal cancer. Magn Reson Med.2020;84: 31. Storz P. Acinar cell plasticity and development of pancreatic ductal adenocarci- 348–364. noma. Nat Rev Gastroenterol Hepatol. 2017;14:296–304. 19. Ariston Gabriel AN, Jiao Q, Yvette U, et al. Differences between KC and KPC 32. Wood LD, Canto MI, Jaffee EM, et al. Pancreatic cancer: pathogenesis, screening, pancreatic ductal adenocarcinoma mice models, in terms of their modeling biol- diagnosis, and treatment. Gastroenterology. 2022;163:386–402.e1. ogy and their clinical relevance. Pancreatology.2020;20:79–88. 33. Alexander DC, Dyrby TB, Nilsson M, et al. Imaging brain microstructure with 20. Bilreiro C, Fernandes FF, Andrade L, et al. Effective bowel motion reduction in diffusion MRI: practicality and applications. NMR Biomed. 2019;32:e3841. mouse abdominal MRI using hyoscine butylbromide. Magn Reson Med.2021; 34. Gholizadeh N, Greer PB, Simpson J, et al. Characterization of prostate cancer 86:2146–2155. using diffusion tensor imaging: a new perspective. Eur J Radiol. 2019;110: 21. Kellner E, Dhital B, Kiselev VG, et al. Gibbs-ringing artifact removal based on lo- 112–120. cal subvoxel-shifts. Magn Reson Med. 2016;76:1574–1581. 35. Toshima F, Watanabe R, Inoue D, et al. Ct abnormalities of the pancreas associated 22. Veraart J, Novikov DS, Christiaens D, et al. Denoising of diffusion MRI using ran- with the subsequent diagnosis of clinical stage I pancreatic ductal adenocarcinoma dom matrix theory. Neuroimage. 2016;142:394–406. more than 1 year later: a case-control study. Am J Roentgenol. 2021;217: 1353–1364. 23. Veraart J, Sijbers J, Sunaert S, et al. Weighted linear least squares estimation of dif- fusion MRI parameters: strengths, limitations, and pitfalls. Neuroimage.2013;81: 36. Sotozono H, Kanki A, Yasokawa K, et al. Value of 3-T MR imaging in intraductal 335–346. papillary mucinous neoplasm with a concomitant invasive carcinoma. Eur Radiol. 2022;32:8276–8284. 24. Harkins K, Does M. GitHub - remmi-toolbox/remmi-matlab: Matlab toolbox to reconstruct and process MRI data. Available at: https://github.com/remmi- 37. Szczepankiewicz F, Lasič S, van Westen D, et al. Quantification of microscopic toolbox/remmi-matlab. Accessed December 2, 2022. diffusion anisotropy disentangles effects of orientation dispersion from micro- structure: applications in healthy volunteers and in brain tumors. Neuroimage. 25. Veite-Schmahl MJ, Rivers AC, Regan DP, et al. The Mouse Model of Pancreatic 2015;104:241–252. Cancer Atlas (MMPCA) for classification of pancreatic cancer lesions: a large his- tological investigation of the Ptf1aCre/+;LSL-KrasG12D/+ transgenic mouse 38. Kenner B, Chari ST, Kelsen D, et al. Artificial intelligence and early detection of model of pancreatic cancer. PloS One. 2017;12:e0187552. pancreatic cancer: 2020 summative review. Pancreas. 2021;50:251–279. 26. Basturk O, Hong SM, Wood LD, et al. A revised classification system and recom- 39. Heredia-Soto V, Rodríguez-Salas N, Feliu J. Liquid biopsy in pancreatic cancer: mendations from the Baltimore consensus meeting for neoplastic precursor le- are we ready to apply it in the clinical practice? Cancers (Basel).2021;13. sions in the pancreas. Am J Surg Pathol. 2015;39:1730–1741. 40. Gurney-Champion OJ, Froeling M, Klaassen R, et al. Minimizing the acquisition 27. Berman-Booty LD, Sargeant AM, Rosol TJ, et al. A review of the existing grading time for intravoxel incoherent motion magnetic resonance imaging acquisitions in schemes and a proposal for a modified grading scheme for prostatic lesions in the liver and pancreas. Invest Radiol.2016;51:211–220. TRAMP mice. Toxicol Pathol.2012;40:5–17. 41. Concia M, Sprinkart AM, Penner A-H, et al. Diffusion-weighted magnetic reso- 28. Knoblaugh SE, Himmel LE. Keeping score: semiquantitative and quantitative nance imaging of the pancreas: diagnostic benefit from an intravoxel incoherent scoring approaches to genetically engineered and xenograft mouse models of can- motion model-based 3 b-value analysis. Invest Radiol.2014;49:93–100. cer. Vet Pathol.2019;56:24–32. 42. Baskaya F, Lemainque T, Klinkhammer B, et al. Pathophysiologic mapping of 29. Bankhead P, Loughrey MB, Fernández JA, et al. QuPath: open source software for chronic liver diseases with longitudinal multiparametric MRI in animal models. digital pathology image analysis. Sci Rep. 2017;7:16878. Invest Radiol. 2024;59:699–710. 10 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Investigative Radiology Wolters Kluwer Health

Pancreatic Intraepithelial Neoplasia Revealed by Diffusion-Tensor MRI

Download PDF
10 pages

Loading next page...
 
/lp/wolters-kluwer-health/pancreatic-intraepithelial-neoplasia-revealed-by-diffusion-tensor-mri-aA0yi49ffC

References (42)

Publisher
Wolters Kluwer Health
Copyright
Copyright © 2024 The Author(s). Published by Wolters Kluwer Health, Inc.
ISSN
1536-0210
eISSN
0020-9996
DOI
10.1097/rli.0000000000001142
Publisher site
See Article on Publisher Site

Abstract

ORIGINAL ARTICLE Pancreatic Intraepithelial Neoplasia Revealed by Diffusion-Tensor MRI Carlos Bilreiro, MD, Francisca F. Fernandes, MSc, Rui V. Simões, PhD, Rafael Henriques, PhD, Cristina Chavarrías, PhD, Andrada Ianus, PhD, Mireia Castillo-Martin, MD, PhD, Tânia Carvalho, DVM, PhD, Celso Matos, MD, and Noam Shemesh, PhD Conclusions: DTI and T2* are useful for detecting and characterizing PanIN in Objectives: Detecting premalignant lesions for pancreatic ductal adenocarci- genetically engineered mice and in the human pancreas, especially with AD and FA. noma, mainly pancreatic intraepithelial neoplasia (PanIN), is critical for early di- These are encouraging findings for future clinical applications of pancreatic imaging. agnosis and for understanding PanIN biology. Based on PanIN's histology, we hy- Key Words: diffusion tensor imaging, carcinoma in situ, pancreatic neoplasms, pothesized that diffusion tensor imaging (DTI) and T2* could detect PanIN. pancreatic intraepithelial neoplasia, animals, genetically modified Materials and Methods: DTI was explored for the detection and characteriza- tion of PanIN in genetically engineered mice (KC, KPC). Following in vivo (Invest Radiol 2024;00: 00–00) DTI, ex vivo ultrahigh-field (16.4 T) MR microscopy using DTI, T2* was per- formed with histological validation. Sources of MR contrasts and histological ancreatic cancer, mainly represented by ductal adenocarcinoma features were investigated, including histological scoring for disease burden (le- P (PDAC), is the third leading cause of cancer-related death in the sion span) and severity (adjusted score). To test if findings in mice can be trans- 1 United States, with an estimated 11% survival rate at 5 years. As symp- lated to humans, human pancreas specimens were imaged. toms develop late in the course of the disease, most patients are diagnosed Results: DTI detected PanIN and pancreatic ductal adenocarcinoma in vivo (6 at advanced stages precluding pancreatic resection surgery, which re- KPC, 4 KC, 6 controls) with high discriminative ability: fractional anisotropy mains the only potentially curative therapeutic option. Moreover, when (FA) and radial diffusivity with area under the curve = 0.983 (95% confidence in- patients are diagnosed in the early disease stages and subjected to tumor 3,4 terval: 0.932–1.000); mean diffusivity and axial diffusivity (AD) with area under resection, the 5-year survival rate reaches 25%–50%. Therefore, the curve = 1 (95% confidence interval: 1.000–1.000). MR microscopy with his- identifying precursor lesions for PDAC, mainly composed of pancreatic tological correlation (20 KC/KPC; 5 controls) revealed that sources of MR con- intraepithelial neoplasia (PanIN), could provide opportunities for early trasts likely arise from microarchitectural signatures: high FA, AD in fibrotic 5,6 diagnosis and development of more effective therapies. However, areas surrounding lesions, high diffusivities within cysts, and high T2* within le- PanINs are not diagnosed by current imaging modalities, with a single sions' stroma. The strongest histological correlations for lesion span and adjusted study describing small cystic lesions in magnetic resonance cholangio- score were obtained with AD (R =0.708, P < 0.001; R = 0.789, P < 0.001, respec- 7 pancreatography in a subset of patients with these lesions. The absence tively). Ex vivo observations in 5 human pancreases matched our findings in of noninvasive diagnostic tools also prevents the investigation of mice, revealing substantial contrast between PanIN and normal pancreas. PanIN's biology and tumorigenesis in humans, which remains largely unknown and mostly driven by research using genetically engineered 8–10 mouse models (GEMMs) and histopathological analyses. There- fore, there is an urgent need for developing imaging methods for PanIN Received for publication September 12, 2024; and accepted for publication, after revi- diagnosis and characterization, which could enable early diagnosis be- sion, October 11, 2024. From the Radiology Department, Champalimaud Foundation, Lisbon, Portugal (C.B., fore PDAC is established. C.M.); Champalimaud Research, Champalimaud Foundation, Lisbon, Portugal Diffusion-weighted MRI (DWI) provides insights into tissue mi- (C.B., F.F.F., R.H., C.C., A.I., M.C.-M., T.C., C.M., N.S.); Nova Medical School, crostructure and is routinely used for diagnosis, staging, and evaluation Lisbon, Portugal (C.B.); i3S—Instituto de Investigação e Inovação em Saúde, of response to therapy in benign and malignant diseases, mostly based Universidade do Porto, Porto, Portugal (R.V.S.); and Pathology Department, 11–14 Champalimaud Foundation, Lisbon, Portugal (M.C.-M.). on apparent diffusion coefficients (ADCs). Given the potential of Conflicts of interest and sources of funding: N.S. serves on the Scientific Advisory DWI, diffusion tensor imaging (DTI) was developed and explored in Board of Bruker Biospin. No other potential conflicts of interest are reported from the clinical setting to provide a rotationally invariant representation of remaining authors. Champalimaud Foundation funded this work; R.V.S.'s work the diffusion process in biological tissues. In the pancreas, DWI has was supported by H2020-MSCA-IF-2018, ref: 844776; A.I.'s work received the support of a fellowship from “La Caixa” Foundation (ID 100010434) and from been used in the past for characterizing benign and malignant pro- the European Union's Horizon 2020 research and innovation programme under cesses, but studies exploring DTI for pancreatic disease characteriza- the Marie Skłodowska-Curie grant agreement no. 847648, fellowship code LCF/ tion remain lacking, including for the detection of PanIN and BQ/PI20/11760029. 15,16 early PDAC. Correspondence to: Noam Shemesh, PhD, Champalimaud Research, Champalimaud Foundation, Av. Brasilia 1400-038, Lisbon, Portugal. E-mail: noam.shemesh@ We hypothesized that high-resolution DTI, with its ability to neuro.fchampalimaud.org. characterize diffusion anisotropy and sensitivity toward microarchitecture, Data availability statement: The data generated in this study are available upon request from could detect and characterize microstructural changes expected in the corresponding author. Any human-derived data are completely anonymized. PanIN and early PDAC. To test this hypothesis, we imaged GEMM de- Supplemental digital contents are available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article veloping PanIN and PDAC in vivo and harnessed ex vivo MR micros- on the journal’s Web site (www.investigativeradiology.com). copy at ultrahigh field and direct histological validation to understand Copyright © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. This is an the sources of contrast. Similar techniques were previously successful open-access article distributed under the terms of the Creative Commons 17,18 in characterizing malignant lymph nodes with high resolution. Attribution-Non Commercial-No Derivatives License 4.0 (CCBY-NC-ND), where it is permissible to download and share the work provided it is properly cited. The Here, we used 2 well-established mouse models of pancreatic cancer: work cannot be changed in any way or used commercially without permission KC, mostly displaying PanIN lesions, and KPC, with complete pene- from the journal. trance of PDAC. Finally, we tested our hypothesis in human ISSN: 1536-0210/24/0000–0000 pancreas samples. DOI: 10.1097/RLI.0000000000001142 Investigative Radiology � Volume 00, Number 00, Month 2024 www.investigativeradiology.com 1 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 are described in Table 2. Anesthesia was induced with 5% isoflurane MATERIALS AND METHODS (Vetflurane; Virbac, France) mixed with oxygen-enriched (28%) air. Ethics Statement Isoflurane dosage was reduced to 1.5%–2.5% during animal preparation and for the duration of MRI scanning. Breathing rate and rectal temper- Animals were handled in agreement with European FELASA ature were monitored and recorded throughout scans using a pillow pres- guidelines. Animal procedures were conducted according to European Di- sure sensor and an optic fiber temperature probe (SA Instruments Inc, rective 2010/63 and preapproved by institutional and national authorities. Stony Brook, USA). A warm-water recirculating pad was used for body The study protocol involving human biological tissue was preapproved temperature control, maintaining rectal temperature between 35.0°C and by the institutional ethics committee. All participants provided signed 37.0°C. Ophthalmic gel (Vidisic Gel; Bausch+Lomb, Canada) was ap- informed consent before tissue collection, and personal identity was plied to the animals' eyes before each acquisition for preventing eye dry- kept private through pseudonymization in the institution's Biobank. ness. All acquisitions began with routine adjustments: center frequency, Genetically Engineered Mouse Models radiofrequency calibration, and automatic shimming. G12D G12D The KC (Pdx1-Cre, LSL-Kras /Ptf1a-Cre, LSL-Kras ) G12D LoxP In Vivo DTI Animals Selection and KPC (Ptf1a-Cre, LSL-Kras ,p53 ) mouse models were KPC mice presented early-onset disease with pancreatic abnor- used: KC mice develop PanIN and progress to PDAC; KPC mice de- malities in anatomical T2-weighted in vivo imaging and physical signs velop PanIN and PDAC with complete penetrance. Healthy mice of malaise. In contrast, KC mice presented pancreatic abnormalities (C57BL/6 J background) were used as controls. All animals' rearing later in their development. Considering these preliminary findings, we was undertaken in a specific pathogen-free facility in a temperature- decided to use young animals from the KPC and older animals from controlled room, with a 12-hour light/dark cycle with ad libitum access the KC transgenic lines for this part of the study: 6 KPC, 4 KC, and 6 to food and water. The animals' cages were enriched with plastic igloos controls (Supplemental Digital Content 1, Table with sample character- and shredded paper. Given the potential of GEMM for developing ad- istics, http://links.lww.com/RLI/A980), with an estimated statistical vanced pancreatic disease, the animals were continuously monitored power of 0.95 (α = 0.05; estimated effect size = 2.2). The predicted for signs of malaise, namely, weight loss/cachexia, fur thinning, and ap- number of KC mice was 6; however, due to unexpected deaths of 2 athetic or aggressive behavior. Whenever these signs were observed, the KC mice before pancreas extraction, these animals could not be used animals were humanely sacrificed. for histopathological analysis and were excluded from the study. Anatomical In Vivo Imaging In Vivo DTI In a first exploratory phase (Fig. 1A), KC (n = 9) and KPC (n = 6) mice were imaged monthly from 4 months old, to assess the de- KPC (n = 6), KC (n = 4), and control (n = 6) mice were imaged in velopment onset of pancreatic abnormalities. Animals were imaged un- a 9.4 T scanner (BioSpec; Bruker, Germany; 40-mm ID linear transmit- til abnormalities were observed, with a T2-weighted turbo spin-echo se- receive volume coil). Intraperitoneal hyoscine butylbromide (5 mg/kg; quence (Table 1), in 1 T and 9.4 T scanners (respectively, Icon and Buscopan; Boehringer Ingelheim, Spain) was administered 8 minutes BioSpec; Bruker, Germany; 40-mm ID linear transmit-receive volume before acquisition for bowel motion reduction. A diffusion-weighted coil) depending on scanner availability: 41 scans were performed in spin-echo echo-planar-imaging was used for imaging the animals' ab- 1 T; 15 scans were performed in 9.4 T scanners. Animal characteristics domen (Table 1). Datasets were analyzed using in-house developed FIGURE 1. Study design, diffusion-weighted multi-gradient-echo pulse sequence diagram, and preprocessing pipeline. A, Study design with major steps performed in chronological order. B, Diffusion-weighted multi-gradient-echo sequence used for the ex vivo MRI acquisition, which includes a small flip- angle slab-selective excitation followed by diffusion gradients and phase encoding, and a multiple-gradient-echo readout echo-train (RF, radiofrequency; SS, slice selection; PE, phase encoding; RO, readout; GDiff, diffusion gradients; δ, diffusion gradient duration; Δ, diffusion time; TE2 = TE3 = TE4). C, Data preprocessing pipeline, with a respective example image obtained in each step (MP-PCA = Marchenko-Pastur principal component analysis). Difference map is the resulting subtraction of pixel values between the raw image and the processed image after denoising and unringing, displaying the removal of image noise and ringing artifacts. 2 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. Investigative Radiology Volume 00, Number 00, Month 2024 DTI Detects Pancreatic Precancerous Lesions TABLE 1. MRI Acquisition Parameters Pulse Sequence T2-TSE SE-EPI DMGE Magnetic field strength 1 T; 9.4 T 9.4 T 16.4 T In-plane resolution (mm ) 0.150  0.150 (1 T) 0.125  0.125 - 0.125  0.125 (9.4 T) Slice thickness (mm) 0.7 (1 T) 0.25 - 0.3 (9.4 T) No. slices 20–36 36 - 3D resolution (mm) - - 0.08  0.08  0.08 TR (ms) 3825 (1 T) 2300–2600 125 1500–2700 (9.4 T) TE (ms) 60 (1 T) 15.6–16.5 9.2–10.2 20–22 (9.4 T) Signal averages 30 (1 T) 16 1 10–16 (9.4 T) Acceleration factor 8 - - Multishot segments - 4 - Partial Fourier reduction factor - 1.7 - EPI sampling bandwidth (kHz) - 395 - Fat suppression No Yes Yes No. DTI B values - 1 1 Bvalues (s/mm ) - 1000 775–1200 No. B = 0 s/mm -2 2 Diffusion directions - 10 10 Respiratory triggering Yes Yes - Scan duration 44–50 min (1 T) 31–33 min 5 h 54 min–16 h 33 min 12–21 min (9.4 T) T2-TSE, T2-weighted turbo spin echo; SE-EPI, spin-echo echo planar imaging; DMGE, diffusion-weighted multi-gradient-echo. code in MATLAB (MathWorks Inc, Natick, MA). Data preprocessing abnormalities were evident in T2-weighted images at 8.5 (interquartile included denoising based on Marchenko-Pastur principal component range [IQR] = 0.75) months of age for KC mice, these animals were 21,22 analysis (MP-PCA) and Gibbs unringing. Fractional anisotropy studied at this age (n = 12), whereas n = 8 KPC mice were studied at (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusiv- 4.25 (IQR = 0.25) months old, right after the first exploratory MRI ity (RD) maps were produced, fitting DTI to diffusion-weighted images scans, due to early aggressive disease. Preliminary findings were used using a weighted-least-squares solution. The images obtained were for calculating the number of animals in each group: pancreases har- analyzed by a gastrointestinal radiologist, who segmented the entire vested from 3 KC mice and 1 healthy control were imaged and FA maps pancreas in each slice, using MATLAB's built-in segmentation tools. were produced. According to a comparison of median FA measure- Measurements obtained from pancreatic segmentation were used for ments, effect size was estimated at 1.6, and a similar effect size was as- groupwise comparisons. After MRI, all animals had their pancreases sumed for KPC mice. With a statistical power of 0.95, each animal harvested and subjected to histopathological diagnosis, so disease status group was initially aimed at 10 animals. However, due to unpredictable (presence of PanIN, PDAC) was known for each animal. events when developing the colonies, 12 KC (available surplus in litter production) and 8 KPC mice were finally used (3 KPC mice suffered premature deaths and their pancreases could not be harvested on time). Pancreas Specimens' Preparation for MR Microscopy GEMM and n = 5 controls were sacrificed with cervical dislocation, Pancreases from additional KPC (n = 8), KC (n = 12), and and their pancreases dissected and removed through median laparot- control (n = 5) mice were prepared for MR microscopy. As pancreatic omy. Pancreatic specimens were subjected to fixation: immersion in TABLE 2. Sample Characteristics in Anatomical Longitudinal T2-Weighted Imaging of GEMM TABLE 3. Lesion Adjusted Score Calculation, Adapted From 25 26 Veite-Schmahl et al and Basturk et al Animal Model KC KPC Score Grade Distribution No. animals 9 6 Pancreatic abnormalities 8.5 (IQR = 0.75) 4.25 (IQR = 0.25) 1 Low-grade Focal onset (months old) 2 High-grade Multifocal Animal weight (g) 30.4 (IQR = 11.2) 22.9 (IQR = 4.1) 3PDAC Diffuse Animal sex 4/9 females 4/6 females The final adjusted score for each pancreatic specimen was determined using the Continuous variables expressed as medians and IQR. Pancreatic abnormalities following distribution-adjusted lesion grade = (low-grade  distribution) + (high- onset was assessed from 4 months old. grade  distribution) + (PDAC  distribution). © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. www.investigativeradiology.com 3 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 4% paraformaldehyde for 48 hours and then in phosphate-buffered sa- distribution) + (PDAC  distribution) with ascribed scores of 1, 2, line with 0.01% sodium azide for 24 hours. After fixation, pancreases and 3 for low-grade PanIN, high-grade PanIN, and PDAC, respectively, were immersed in Fluorinert within a 10-mm NMR tube for ex vivo and for focal, multifocal, and diffuse lesions, respectively. The same MRI, with special care for removing air bubbles and animal hair in histological slide was also used to calculate lesion span (ie, lesion bur- the sample. den for each specimen), through digital segmentation using QuPath v. 0.4.3 semimanual segmentation tools, by annotating area of the lesions 28,29 Ex Vivo MR Microscopy and area of normal pancreas. Pancreases were imaged in a 16.4 T MRI scanner (Ascend Aeon; Lesion span and adjusted score were correlated with MRI data, Bruker, Germany), using a diffusion-weighted multi-gradient-echo se- assessing if DTI and T2* measurements scale with disease extension quence, providing ultrahigh-resolution 3D images with both diffusion and severity. and T2* weighting (Fig. 1B, Table 1). Two 10-mm probes were used (Micro5, Cryoprobe; Bruker, Germany) depending on availability; sam- Imaging and Histology Alignment ple temperature was maintained at 37°C. Datasets were preprocessed as The alignment between MRI and histology was manually per- previously described (MP-PCA denoising, Gibbs unringing, 21,22 formed by a gastrointestinal radiologist. For this, the Volume Viewer Fig. 1C). After preprocessing, images were masked to remove tool from ImageJ (US National Institutes of Health) was used for low-signal areas. T2*, DTI (FA, MD, AD, RD), and ADC maps were reconstructing the 3D datasets in real time and searching for image then obtained from the preprocessed datasets. planes corresponding to histology slides using the powder-averaged dif- Histological Analysis fusion-weighted images. Specimens' structures (lymph nodes, large blood vessels, cystic lesions, and large masses) were used as landmarks Following MRI, mouse pancreas samples were serially sectioned (4-μm thickness, 0.5-mm spacing) and stained with hematoxylin and for searching for the most adequate image plane. When the manually re- eosin. Slides were digitally scanned in Ultra-Fast Scanner (Philips) or constructed image plane and slice were deemed the most approximate AxioScan Z1 (Zeiss) and examined by a veterinary pathologist. as possible to the corresponding histological slide, the reconstruction The classification of pancreatic lesions—acinar-to-ductal meta- coordinates were used in all DTI and T2* maps, and the corresponding plasia (ADM), low-grade PanIN, high-grade PanIN, and PDAC—was images were exported for a visual comparison between both histology based on specific criteria (Supplemental Digital Content 2, Table with and MRI. Image contrasts were therefore attributed to the correspond- classification, http://links.lww.com/RLI/A980), adapted from Veite- ing histological changes, as directly observed. 25 26 Schmahl et al and Basturk et al. Disease severity scoring was per- formed on the slide where the extent of the lesions was higher. A Human Pancreas distribution-adjusted lesion grade with equal contribution from each le- sion type was developed, ensuring that all lesions contribute equally to Pancreas specimens (n = 5) obtained from patients with PDAC, the final score, rendering a continuous range of scores, adapted from from the institution's Biobank (formalin-fixed leftovers after diagnostic pro- 27 28 30 Berman-Booty et al and Knoblaugh and Himmel. This system cal- cedures following resection surgery), were analyzed searching for PanIN. culates the score by summing the product of the scores for grade and Specimens were sliced and placed inside a 10-mm MR tube. Sample prep- distribution of each lesion, as shown in Table 3: distribution-adjusted le- aration, imaging acquisition, and processing methods described above for sion grade = (low-grade PanIN  distribution) + (high-grade PanIN- MR microscopy were used. Histological slides were sectionized, stained, FIGURE 2. Pancreatic abnormalities with cystic changes are seen in vivo in KPC (4.75 months old) and KC (8.5 months old) mice, using conventional TSE- T2-weighted images, with corresponding histological diagnosis, including acinar-to-ductal metaplasia, PanIN, and PDAC lesions. The healthy control (2 months old) pancreas is homogeneous without focal or diffuse changes. C, A necrotic/cystic area inside a large tumor (arrows) is directly correlated with histology. All 3 images were acquired in a 9.4 T scanner (BioSpec). 4 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. Investigative Radiology Volume 00, Number 00, Month 2024 DTI Detects Pancreatic Precancerous Lesions and digitized, as described above, and evaluated by a gastrointestinal pa- RESULTS thologist. Alignment was performed manually by a gastrointestinal radiol- ogist, as previously described, for a visual comparison. Longitudinal MRI Assessment of PanIN and PDAC Development Onset Exploratory anatomical in vivo T2-weighted images revealed Statistical Analysis variably enlarged pancreases with high signal intensity and cystic le- IBM SPSS statistics v.23 was used. Kolmogorov-Smirnov test was sions for GEMM, whereas the pancreas of control mice was homoge- used to assess the distribution of continuous variables, revealing nonnormal neous with low signal intensity (Fig. 2A). These findings were evident distributions throughout. Kruskal-Wallis test was used for groupwise com- at 8.5 months old for KC mice and at first scans (4.25 months old) for parisons, with post hoc pairwise comparisons, and Mann-Whitney U test KPC mice (Table 2), as confirmed with histopathology (Fig. 2B). for comparisons between 2 groups. Bonferroni correction was applied when comparing multiple test metrics, namely, different DTI, T2*, and DTI Noninvasively Detects PanIN and PDAC In Vivo ADC maps, by multiplying the P values by the number of metrics. Re- ceiver operating characteristics analysis was performed for assessing the Both KC and KPC mice presented pancreases with high FA, discriminative value of DTI. Pearson correlation coefficient was used for MD, AD, and RD; large PDAC; and small cystic lesions, contrasting vi- evaluating MRI measurements and lesion span/adjusted score. All tests sually with control mice which presented low values in the pancreas in were bilateral; P values <0.05 were considered significant. all parameters (Fig. 3A). FIGURE 3. DWIs (averaged b = 0 and b = 1000 s/mm images) and DTI maps (FA, MD, AD, RD) for representative KC, KPC, and healthy control mice. A, The maps produced with DTI showcase the differences between healthy controls, KC, and KPC mice. The KC presented here, with diffuse PanIN infiltration in the pancreatic parenchyma and no PDAC, has visually increased values in DTI derived maps, especially FA and AD. The KPC mouse is bearing a large PDAC, also with increased values in the same maps. In contrast, the healthy control mouse has a homogeneous pancreas with intermediate-low values in all maps. B, Quantitatively, the differences between groups of animals, including 10 GEMM (6 KPC and 4 KC) and 6 controls, provide a good discriminative ability of DTI-derived metrics. FA presents the best performance when using median values (AUC = 0.983); AD and MD present the best performance when using 95th percentile values (AUC = 1). MD, AD, RD expressed in mm /s. © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. www.investigativeradiology.com 5 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 GEMM and controls (Fig. 5; Supplemental Digital Content 4, Figure TABLE 4. In Vivo DTI With KC and KPC Mice—Presence of PDAC with all group comparisons, http://links.lww.com/RLI/A981). When comparing KPC and controls, median values were significantly different, PDAC Present Absent P with KPC presenting higher values for MD (+10.6%, P = 0.004), AD (+22%, P = 0.001), and RD (+7.7%, P = 0.023). When comparing KC No. animals 7 3 - and controls, only median AD was different (+14%, P = 0.046). The com- Models KC: 1/4 KC: 3/4 0.033* parison of highest values in each specimen (95th percentile) enhanced the KPC: 6/6 KPC: 0 observed differences between GEMM and controls. For KPC versus con- FA (median) 0.25 0.28 0.833 trols, except for FA (P = 0.128), KPC presented higher MD (+41.2%, FA (95th percentile) 0.45 0.48 0.383 P = 0.006), AD (+45.4%, P = 0.0018), RD (+38.9%, P = 0.004), T2* MD (median) 1.6E-3 1.5E-3 1.000 (+69.6%, P = 0.017), and ADC (+43.2%, P = 0.005). For KC versus con- MD (95th percentile) 2.1E-3 2.0E-3 0.517 trols, KC presented higher FA (+186.7%, P = 0.003), MD (+38.2%, AD (median) 2.1E-3 1.9E-3 0.517 P = 0.006), AD (+47.5%, P = 0.002), RD (+35.7%, P = 0.015), T2* AD (95th percentile) 2.5E-3 2.5E-3 0.833 (+58.6%, P = 0.010), and ADC (+40.8%, P = 0.006). When comparing RD (median) 1.3E-3 1.3E-3 1.000 KC and KPC, no differences were observed for median or 95th percentile RD (95th percentile) 1.9E-3 1.8E-3 0.517 values; however, the lowest values (5th percentile) were lower in KC for MD (−15.8%, P = 0.003), AD (−14.4%, P = 0.007), and RD (−15.2%, Continuous variables were expressed as medians. P = 0.005). Variable median ADC for different diffusion directions in each Fisher exact test was used for comparing prevalence of PDAC in GEMM animal further demonstrates diffusion anisotropy in the pancreas (Supple- models. mental Digital Content 5, Figure with median ADC for all diffusion Mann-Whitney U test was used for comparing GEMM with and without directions, http://links.lww.com/RLI/A982). PDAC for each DTI metric. Histologically, ADM was present in all GEMM mice, most *Statistically significant results. markedly in KC mice, as shown in Supplemental Digital Content 6 (Figure, representative segmented samples, http://links.lww.com/RLI/ A983) and Table 5 (all samples). PDAC was mainly observed in KPC Median FA obtained after pancreatic segmentation was signifi- mice: 33.33% of KC and 87.5% of KPC mice. Both GEMM had PanIN cantly higher for both transgenic mice (n = 6 KPC, n = 4 KC) compared lesions of grades 1 to 3, more marked in KC mice. Most mice had with controls (n = 6), with differences increasing and becoming signifi- ≥70% of their pancreas affected: 75% of KC mice and 75% of KPC cant also for MD, AD, and RD, when assessing the higher values ob- mice. The median lesion span in KC mice was 80.88% (IQR: 42.75) served in each group (ie, 95th percentile) (Fig. 3B). The discriminative and in KPC mice was 91.45% (IQR: 36.62). Median adjusted score performance of median DTI measurements obtained from the entire pan- was 17 (IQR: 9) in KC mice and 26 (IQR: 1.75) in KPC mice. creas in each animal for differentiating GEMM from controls was high, ranging from 0.767 (95% confidence interval [CI]: 0.527–1.000) area un- der the curve (AUC) for RD, up to 0.983 (95% CI: 0.932–1.000) AUC for DTI and T2* Correlate With Disease Burden FA. The discriminative performance increased once more when assessing and Severity the parameters' 95th percentile, with AUC ranging from 0.983 (FA and As seen in Figure 5, Supplemental Digital Content 7 (http:// RD, 95% CI: 0.932–1.000) to 1.000 (MD and AD, 95% CI: links.lww.com/RLI/A984), 8 (http://links.lww.com/RLI/A985) (Fig- 1.000–1.000) (Fig. 3B). When comparing KC and KPC mice, the DTI ures, lesion span and adjusted score correlation plots), and 9 (Table, le- metrics were similar between both groups of animals. Also, DTI metrics sion span and adjusted score correlation coefficients, http://links.lww. could not distinguish between GEMM with (n = 7) and without PDAC com/RLI/A980), DTI/T2* metrics correlate both with disease burden (n = 3) (Table 4). (lesion span) and severity (adjusted score), when using median and 95th percentile values. The highest correlation coefficients both for le- MR Microscopy With DTI and T2* Characterizes Ex Vivo sion span and for adjusted score were obtained with AD (median AD, R = 0.789; 95th percentile AD, R = 0.729; respectively). Other than Pancreatic Abnormalities FA and T2*, 5th percentile values of MRI metrics were not correlated To better understand the contrasts observed, we performed de- with histological scores. tailed MR microscopy and histology in pancreases harvested from KC mice (n = 12), KPC mice (n = 8), and healthy control mice (n = 5) PanIN in the Human Pancreas Revealed by Ex Vivo DTI (Supplemental Digital Content 3, Table with sample characteristics, http:// links.lww.com/RLI/A980). Human pancreas samples (Supplemental Digital Content 10, Ta- After alignment between MR microscopy and histology slides, ble with patient characteristics, http://links.lww.com/RLI/A980) pre- enabling a direct comparison of MRI and histology, stark contrasts be- sented relevant histological differences when compared with the tween DTI and T2* parameters in normal and abnormal tissue were ob- GEMM counterparts: scattered foci of ADM, higher proportion of nor- served (Fig. 4). High FA values are observed in areas of ADM, where mal pancreatic parenchyma, and absence of PDAC (Fig. 6, Supplemen- PanIN lesions of variable degrees are found surrounded by fibrosis, tal Digital Content 11, http://links.lww.com/RLI/A986). In 3 samples, as well as at the periphery of PDAC lesions. In contrast, areas of normal parenchymal atrophy with fatty replacement was observed (Figs. 6B parenchyma demonstrate low FAvalues, allowing a clear distinction be- and C, Supplemental Digital Content 11B, http://links.lww.com/RLI/ tween different types of tissue. The MD, AD, RD, and ADC maps reveal A986). After alignment between MR microscopy and histology, we ob- high values in cystic and necrotic areas, most evident in AD maps, and served clear contrasts between areas of ADM, PanIN, and surrounding intermediate/high AD values in fibrotic areas of ADM, PanIN, and normal pancreas. High FA, AD, and T2* values were observed in areas PDAC. T2* maps present high values in the stroma of both PanIN of ADM and surrounding PanIN lesions; contrast between lesions and and PDAC lesions. All the control pancreases show homogeneous surrounding parenchyma was evident in all maps. One pancreas sample FA, diffusivities, T2*, and ADC maps with lower values throughout with diffuse interstitial fibrosis, attributed to chronic pancreatitis, the pancreatic parenchyma. showed mid-high FA, AD, and T2* values irregularly scattered through- Quantitatively, as observed for in vivo DTI, DTI/T2* maps ob- out the fibrotic tissue (Fig. 6C). Histologically, this diffuse interstitial fi- tained from pancreas specimens allowed us to distinguish between brosis is haphazardly organized, whereas the periductal fibrosis forms a 6 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. Investigative Radiology Volume 00, Number 00, Month 2024 DTI Detects Pancreatic Precancerous Lesions FIGURE 4. Comparison between MR microscopy and histology in GEMM. A, The healthy mouse pancreas is homogeneous, with low anisotropy, diffusivity, and T2* values; coil used: Cryoprobe. B, KC pancreas with extensive ADM and PanIN (black arrows), as well as PDAC (white arrows). Conventional diffusion weighting shows heterogeneous restricted diffusion. There are high anisotropy values in the periphery of areas with ADM, PanIN, and PDAC. High axial diffusivity is observed in cystic/necrotic areas, and intermediate diffusivity and high T2* values are seen in the lesions' stroma. Coil used: Cryoprobe. C, KPC pancreas with extensive PDAC (white arrows) and small areas of ADM and PanIN (black arrows). Conventional diffusion weighting is not helpful in delineating the pancreatic lesions. There are high anisotropy values in the periphery of PDAC and PanIN lesions; high diffusivity values in cystic/necrotic areas and regions with inflammatory infiltrates; and high T2* values in the lesions' stroma. Coil used: Micro5. AD is expressed in mm /s; T2* is expressed in milliseconds (ms). sheath or cuff-like structure around ducts in areas of ADM, PanIN, Here, we leverage GEMM for DTI detection of PanIN, PDAC, and and PDAC. associated pancreatic changes, suggesting DTI/T2* as potential bio- markers for disease onset, staging, or monitoring. Our results also high- DISCUSSION light the important variability between pancreatic structural changes in these models, probably caused by variable proportions of tumoral This study provides evidence establishing a potential role for stroma, ductal structures, and inflammatory infiltrates, which should DTI and T2* in characterizing PanIN and PDAC. We report that ex vivo be accounted for in future studies. and in vivo MRI detects PanIN-associated changes and PDAC, identi- The DTI and T2* maps provided more striking visual contrasts fying potential mechanisms for the contrast and underscoring their than conventional diffusion-weighted images, the method currently translational relevance for imaging the human pancreas. used in clinical practice. As observed by comparing MRI with histol- The process of carcinogenesis in PDAC follows a progressive ogy, these contrasts appear to be derived from specific microstructural development of PanIN, where genetic mutations and morphological 10,31 characteristics of the lesions: areas of anisotropy with high FA values changes are acquired until PDAC is finally established. There is a corresponded to the peripheral zones of PanIN, ADM, and PDAC, recognized absence of noninvasive detection methods for PanIN, pre- 7,32 where sheath/cuff-like periductal fibrosis is observed; areas of high dif- cluding research into PanIN biology, especially in humans. Most fusivities corresponded mostly to cystic and necrotic regions; areas of current knowledge of PanIN was achieved with studies performed in 9,31 high T2* corresponded mostly to stromal components. Also, AD per- GEMM. Thus, the development of noninvasive detection methods formed better than other metrics, including the clinical standard ADC, can not only inform about PanIN biology and tumorigenesis, but also for lesion detection and correlated more closely with disease extent provide opportunities for early therapeutic interventions or close and severity. These results hint to differences in tissue microstructure follow-up of patients, enabling PDAC diagnosis in resectable stages. © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. www.investigativeradiology.com 7 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 FIGURE 5. Group comparisons, lesion span, and adjusted score for DTI (FA and AD are shown) and T2* metrics. For group comparisons, the differences between GEMM and controls are highlighted when using the highest (95th percentile) values, where AD provided the largest differences. For lesion span correlation, the use of 95th percentile values improved the results, where AD was the most correlated metric. For adjusted score, both medians and the 95th percentile values showed significant correlations with disease burden and severity, but the 95th percentile values did not show an improvement when compared with median values. Again, AD provided the best performance (R = 0.622, when using the median values). The group comparisons, lesion span, and adjusted score correlations for all MRI metrics can be found in Supplemental Digital Content 4, 7, and 8. as the origin of the observed MRI contrasts and highlight the value of sented with increased size and parenchymal replacement by abnormal microstructural characterization through DTI, as previously described tissue. The presence of fatty tissue is not observed in our DTI/T2* data, 13,33,34 for other applications. as the acquisitions were performed with fat suppression, and no fat- PanIN remains an elusive lesion in humans. Noncommunicating sensitive pulse sequences were performed. The characterization of fatty pancreatic microcysts in magnetic resonance cholangiopancreatogra- replacement is an interesting point for future studies, especially if per- phy were reported to predict the presence of PanIN in pancreatic spec- formed on the human pancreas. imens, with 61% accuracy. Our findings show that DTI/T2* might be DTI has been performed successfully for imaging the pancreas useful in this context, as the contrasts seen in human pancreas allowed a in clinical context. However, its clinical role remains undetermined, clear detection of PanIN. Regarding differentiation between degrees of both for PDAC detection/characterization and for characterizing other PanIN and PDAC, we found a clear correlation between disease burden malignant lesions or nonmalignant pancreatic processes. Technical dif- and severity and DTI/T2* metrics; however, the visual contrasts were ferences between MRI in basic research and clinical imaging are obvi- mostly similar, both in vivo and ex vivo. This may question the specific- ous, and pulse sequence adaptations with loss of resolution are expected ity of DTI/T2* for visually differentiating low-grade from high-grade due to hardware constraints in clinical scanners and time constraints PanIN and PDAC, but further data are needed to address this question. when imaging patients. For PanIN detection, information loss with The presence of parenchymal atrophy with fatty replacement has lower resolutions in 1.5 T and 3 T scanners might be overcome with been reported previously, as a finding associated with PanIN and early the use of pulse sequences sensitive to microscopic anisotropy. Future 7,35,36 PDAC. Our findings in human pancreas specimens corroborate studies should investigate if DTI can be used in clinical context for these reports, with 3 out of 5 samples presenting fatty replacement. This PanIN detection and characterization, addressing the question of speci- was not observed on GEMM, however, as the animals' pancreases pre- ficity for high- and low-grade PanIN and PDAC, as this issue is most TABLE 5. Histological Analysis of KC, KPC, and Control Mice Used for MR Microscopy Animal Model Controls KC KPC ADM 0% 100% (12/12) 100% (8/8) Low-grade PanIN 0% 100% (12/12) 100% (8/8) High-grade PanIN 0% 100% (12/12) 100% (8/8) PDAC 0% 33.33% (4/12) 87.5% (11/12) Lesion span (%) 0% 80.88 (IQR = 42.75) 91.45 (IQR = 36.62) Adjusted score 0 17 (IQR = 9) 26 (IQR = 1.75) Low-grade PanIN includes PanIN 1 and PanIN 2. Continuous variables expressed as medians and IQR. 8 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. Investigative Radiology Volume 00, Number 00, Month 2024 DTI Detects Pancreatic Precancerous Lesions FIGURE 6. The contrasts observed with MR microscopy DTI in the human pancreas are similar to the ones seen in GEMM (2 remaining specimens are presented in Supplemental Digital Content 11). A, High anisotropy, high T2*, and low AD are seen in the periphery of a high-grade PanIN lesion, contrasting well with the surrounding normal pancreatic parenchyma (arrows). B, As observed previously in the KC/KPC mice, areas with ADM and low- grade PanIN have high anisotropy and AD values at their periphery, with high T2* values in the stroma (arrows). Large areas of fatty tissue are seen in the specimen. C, This specimen has extensive haphazardously deposited fibrotic deposition and fatty infiltration, consistent with chronic pancreatitis. There are low-grade PanIN with high anisotropy, intermediate AD, and T2* values (arrows), but the fibrosis has heterogeneous anisotropy, higher AD, and T2* values. Some areas of this specimen have irregular T2* values, probably due to fixation artifacts. AD is expressed in mm /s; T2* is expressed in milliseconds (ms). relevant for implementing therapeutic interventions. Eventual combina- ORCID ID tions of DTI with other diagnostic tools could be used for increasing spec- Noam Shemesh https://orcid.org/0000-0001-6681-5876 38,39 ificity, such as liquid biopsy and artificial intelligence. Also, it would be interesting to investigate if DTI could be used in conjunction with 3D REFERENCES histological mapping, for detecting specific genomic profiles of PanIN 1. Siegel RL, Miller KD, Fuchs HE, et al. Cancer statistics, 2022. CA Cancer J Clin. and PDAC. Multiparametric MRI could also be studied for characterizing 2022;72:7–33. premalignant lesions, harnessing information from different pulse se- 2. Ducreux M, Cuhna AS, Caramella C, et al. Cancer of the pancreas: ESMO clinical quences and including diffusion-based techniques like intravoxel incoher- practice guidelines for diagnosis, treatment and follow-up. Ann Oncol.2015;26 40–42 (Suppl 5):v56–v68. ent motion, which have already demonstrated feasibility. 3. Campbell F, Smith RA, Whelan P, et al. Classification of R1 resections for pancre- We acknowledge some limitations of this research work. First, atic cancer: the prognostic relevance of tumour involvement within 1 mm of a re- GEMM do not fully recapitulate morphological and biological features section margin. Histopathology. 2009;55:277–283. of human PanIN and PDAC, as PanIN lesions in the mouse are mostly 4. Springfeld C, Ferrone CR, Katz MHG, et al. Neoadjuvant therapy for pancreatic present in the context of extensive ADM. Nevertheless, the observed cancer. Nat Rev Clin Oncol. 2023;20:318–337. image contrasts are useful for in vivo animal imaging, and large similar- 5. Ryan DP, Hong TS, Bardeesy N. Pancreatic adenocarcinoma. NEngl J Med. 2014;371:1039–1049. ities with human pancreas samples were observed, suggesting relevance 6. Overbeek KA, Goggins MG, Dbouk M, et al. Timeline of development of pancre- for human imaging. Second, the manual alignment between MRI and atic cancer and implications for successful early detection in high-risk individuals. histology was suboptimal and could be considered a potential source Gastroenterology. 2022;162:772–785.e4. of errors, as automatic registration was not viable due to the expected 7. Vullierme M-P, Menassa L, Couvelard A, et al. Non-branched microcysts of the distortion and configurational changes during sample processing for pancreas on MR imaging of patients with pancreatic tumors who had pancreatec- histology. Third, the in vivo application developed here demonstrates tomy may predict the presence of pancreatic intraepithelial neoplasia (PanIN): a the feasibility of the technique and its usefulness for differentiating preliminary study. Eur Radiol. 2019;29:5731–5741. GEMM from healthy controls but did not evaluate its role in determin- 8. Braxton AM, Kiemen AL, Grahn MP, et al. 3D genomic mapping reveals multifocality of human pancreatic precancers. Nature. 2024. Published online ing different degrees of disease progression in vivo. Also, KC and KPC May 1, 2024:;629:679–687. mice were observed in different age groups due to their different rates of 9. Alonso-Curbelo D, Ho YJ, Burdziak C, et al. A gene-environment-induced epige- disease progression. Future studies evaluating animals of the same netic program initiates tumorigenesis. Nature. 2021;590:642–648. transgenic line at different stages of disease progression should eluci- 10. Storz P, Crawford HC. Carcinogenesis of pancreatic ductal adenocarcinoma. Gas- date if DTI is useful for determining the stage of disease in similar troenterology. 2020;158:2072–2081. age groups. Finally, although similarities of imaging contrasts were ob- 11. Nilsson M, Englund E, Szczepankiewicz F, et al. Imaging brain tumour micro- served when imaging human pancreas samples, it remains to be deter- structure. Neuroimage.2018;182:232–250. mined if DTI can be successfully adapted to clinical scanners for imag- 12. Koh DM, Collins DJ. Diffusion-weighted MRI in the body: applications and chal- lenges in oncology. AJR Am J Roentgenol. 2007;188:1622–1635. ing patients and maintain sensitivity toward these lesions. Our work 13. Kiselev VG, Körzdörfer G, Gall P. Toward quantification: microstructure and provides exciting data in this regard and is encouraging for the develop- magnetic resonance fingerprinting. Invest Radiol.2021;56:1–9. ment of clinical DTI applications for pancreatic imaging. 14. Fukukura Y, Kanki A. Quantitative magnetic resonance imaging for the pancreas. In conclusion, PanIN lesions were detected with DTI and T2* Invest Radiol.2024;59:69–77. contrast in GEMM and human pancreas, suggesting high relevance as 15. Lanzman RS, Wittsack HJ. Diffusion tensor imaging in abdominal organs. noninvasive biomarkers. This augurs well for characterizing PanIN pro- NMR Biomed.2017;30. gression and biology and associated pancreatic changes in both mice 16. Nissan N, Golan T, Furman-Haran E, et al. Diffusion tensor magnetic resonance and humans. imaging of the pancreas. PloS One. 2014;9:e115783. © 2024 The Author(s). Published by Wolters Kluwer Health, Inc. www.investigativeradiology.com 9 Bilreiro et al Investigative Radiology Volume 00, Number 00, Month 2024 17. Santiago I, Santinha J, Ianus A, et al. Susceptibility perturbation MRI maps tumor 30. Andea A, Sarkar F, Adsay VN. Clinicopathological correlates of pancreatic intra- infiltration into mesorectal lymph nodes. Cancer Res. 2019;79:2435–2444. epithelial neoplasia: a comparative analysis of 82 cases with and 152 cases without pancreatic ductal adenocarcinoma. Mod Pathol. 2003;16:996–1006. 18. Ianuş A, Santiago I, Galzerano A, et al. Higher-order diffusion MRI characteriza- tion of mesorectal lymph nodes in rectal cancer. Magn Reson Med.2020;84: 31. Storz P. Acinar cell plasticity and development of pancreatic ductal adenocarci- 348–364. noma. Nat Rev Gastroenterol Hepatol. 2017;14:296–304. 19. Ariston Gabriel AN, Jiao Q, Yvette U, et al. Differences between KC and KPC 32. Wood LD, Canto MI, Jaffee EM, et al. Pancreatic cancer: pathogenesis, screening, pancreatic ductal adenocarcinoma mice models, in terms of their modeling biol- diagnosis, and treatment. Gastroenterology. 2022;163:386–402.e1. ogy and their clinical relevance. Pancreatology.2020;20:79–88. 33. Alexander DC, Dyrby TB, Nilsson M, et al. Imaging brain microstructure with 20. Bilreiro C, Fernandes FF, Andrade L, et al. Effective bowel motion reduction in diffusion MRI: practicality and applications. NMR Biomed. 2019;32:e3841. mouse abdominal MRI using hyoscine butylbromide. Magn Reson Med.2021; 34. Gholizadeh N, Greer PB, Simpson J, et al. Characterization of prostate cancer 86:2146–2155. using diffusion tensor imaging: a new perspective. Eur J Radiol. 2019;110: 21. Kellner E, Dhital B, Kiselev VG, et al. Gibbs-ringing artifact removal based on lo- 112–120. cal subvoxel-shifts. Magn Reson Med. 2016;76:1574–1581. 35. Toshima F, Watanabe R, Inoue D, et al. Ct abnormalities of the pancreas associated 22. Veraart J, Novikov DS, Christiaens D, et al. Denoising of diffusion MRI using ran- with the subsequent diagnosis of clinical stage I pancreatic ductal adenocarcinoma dom matrix theory. Neuroimage. 2016;142:394–406. more than 1 year later: a case-control study. Am J Roentgenol. 2021;217: 1353–1364. 23. Veraart J, Sijbers J, Sunaert S, et al. Weighted linear least squares estimation of dif- fusion MRI parameters: strengths, limitations, and pitfalls. Neuroimage.2013;81: 36. Sotozono H, Kanki A, Yasokawa K, et al. Value of 3-T MR imaging in intraductal 335–346. papillary mucinous neoplasm with a concomitant invasive carcinoma. Eur Radiol. 2022;32:8276–8284. 24. Harkins K, Does M. GitHub - remmi-toolbox/remmi-matlab: Matlab toolbox to reconstruct and process MRI data. Available at: https://github.com/remmi- 37. Szczepankiewicz F, Lasič S, van Westen D, et al. Quantification of microscopic toolbox/remmi-matlab. Accessed December 2, 2022. diffusion anisotropy disentangles effects of orientation dispersion from micro- structure: applications in healthy volunteers and in brain tumors. Neuroimage. 25. Veite-Schmahl MJ, Rivers AC, Regan DP, et al. The Mouse Model of Pancreatic 2015;104:241–252. Cancer Atlas (MMPCA) for classification of pancreatic cancer lesions: a large his- tological investigation of the Ptf1aCre/+;LSL-KrasG12D/+ transgenic mouse 38. Kenner B, Chari ST, Kelsen D, et al. Artificial intelligence and early detection of model of pancreatic cancer. PloS One. 2017;12:e0187552. pancreatic cancer: 2020 summative review. Pancreas. 2021;50:251–279. 26. Basturk O, Hong SM, Wood LD, et al. A revised classification system and recom- 39. Heredia-Soto V, Rodríguez-Salas N, Feliu J. Liquid biopsy in pancreatic cancer: mendations from the Baltimore consensus meeting for neoplastic precursor le- are we ready to apply it in the clinical practice? Cancers (Basel).2021;13. sions in the pancreas. Am J Surg Pathol. 2015;39:1730–1741. 40. Gurney-Champion OJ, Froeling M, Klaassen R, et al. Minimizing the acquisition 27. Berman-Booty LD, Sargeant AM, Rosol TJ, et al. A review of the existing grading time for intravoxel incoherent motion magnetic resonance imaging acquisitions in schemes and a proposal for a modified grading scheme for prostatic lesions in the liver and pancreas. Invest Radiol.2016;51:211–220. TRAMP mice. Toxicol Pathol.2012;40:5–17. 41. Concia M, Sprinkart AM, Penner A-H, et al. Diffusion-weighted magnetic reso- 28. Knoblaugh SE, Himmel LE. Keeping score: semiquantitative and quantitative nance imaging of the pancreas: diagnostic benefit from an intravoxel incoherent scoring approaches to genetically engineered and xenograft mouse models of can- motion model-based 3 b-value analysis. Invest Radiol.2014;49:93–100. cer. Vet Pathol.2019;56:24–32. 42. Baskaya F, Lemainque T, Klinkhammer B, et al. Pathophysiologic mapping of 29. Bankhead P, Loughrey MB, Fernández JA, et al. QuPath: open source software for chronic liver diseases with longitudinal multiparametric MRI in animal models. digital pathology image analysis. Sci Rep. 2017;7:16878. Invest Radiol. 2024;59:699–710. 10 www.investigativeradiology.com © 2024 The Author(s). Published by Wolters Kluwer Health, Inc.

Journal

Investigative Radiology – Wolters Kluwer Health

Published: Dec 13, 2024

There are no references for this article.