Design and Rationale for the Use of Magnetic Resonance Imaging Biomarkers to Predict Diabetes Following Acute Pancreatitis in the Diabetes RElated to Acute Pancreatitis and Its Mechanisms (DREAM) Study: From the Type 1 Diabetes in Acute Pancreatitis Consortium (T1DAPC)

Abstract

This core component of the Diabetes RElated to Acute pancreatitis and its Mechanisms (DREAM) study will examine the hypothesis that advanced magnetic resonance imaging (MRI) techniques can reflect underlying pathophysiologic changes and provide imaging biomarkers that predict diabetes mellitus (DM) following acute pancreatitis (AP). A subset of participants in the DREAM study will enroll and undergo serial MRI examinations using a specific research protocol. We aim to differentiate at-risk individuals from those who remain euglycemic by identifying parenchymal features following AP. Performing longitudinal MRI will enable us to observe and understand the natural history of post-AP DM. We will compare MRI parameters obtained by interrogating tissue properties in euglycemic, prediabetic and incident diabetes subjects and correlate them with metabolic, genetic, and immunological phenotypes. Differentiating imaging parameters will be combined to develop a quantitative composite risk score. This composite risk score will potentially have the ability to monitor the risk of DM in clinical practice or trials. We will use artificial intelligence, specifically deep learning, algorithms to optimize the predictive ability of MRI. In addition to the research MRI, the DREAM study will also correlate clinical computerized tomography and MRI scans with DM development.

INTRODUCTION

According to recent reports, pancreatogenic diabetes mellitus (DM), a known complication of acute pancreatitis (AP), may occur more frequently than previously recognized.¹ However, there are limited prospective data from well-phenotyped patients following AP to confirm these retrospective data. The Type 1 Diabetes in Acute Pancreatitis Consortium (T1DAPC) was recently formed to fill this knowledge gap.² The primary longitudinal study of the T1DAPC is the Diabetes RElated to Acute pancreatitis and its Mechanisms (DREAM) study, which is described elsewhere in this issue. The DREAM study will collect and analyze computed tomography (CT) and magnetic resonance imaging (MRI) scans performed as part of the clinical care of enrolled pancreatitis patients. In addition, we will also conduct research MRI scans in a sub study of DREAM subjects entitled Imaging Morphology of Pancreas in Diabetic Patients Following Acute Pancreatitis (IMMINENT) which is funded separately by the National Institutes of Health (RFA-DK-21-501). The main hypothesis is that MRI can predict the development of prediabetes (pre-DM) or type 1 diabetes mellitus (T1D) and other forms of DM by identifying pathophysiologic changes in the pancreatic parenchyma following AP. We will explore the potential of MRI examinations to differentiate at-risk patients from those who will remain euglycemic. Serial MR examinations will monitor several morphological and pathophysiologic changes. The primary objective is to use several novel MRI techniques to define the natural history of the development of pre-DM and DM in post-AP patients. Secondary objectives include correlating these MRI and artificial intelligence (AI) parameters with the functional, metabolic, genetic, and immunological phenotypes established by the DREAM study.

Background and Rationale

It is well accepted that MRI is superior to CT in demonstrating soft-tissue contrast.³ Magnetic resonance imaging also offers advanced imaging techniques that can be used to interrogate specific tissue properties, such as dynamic contrast-enhanced (DCE) MRI, intravoxel incoherent motion (IVIM), or diffusion-weighted imaging (DWI). Some MRI-detectable phenomena are thought to be involved in the pathogenesis of AP and DM, whereas others may be the consequence of pancreatic disease.

Previous MRI studies have described altered pancreas size and morphology in patients with T1D, even prior to the onset of clinical DM.^4-6 The dynamics of decline in pancreas volume associated with T1D are uncertain; however, we expect the pancreas to be smaller in patients with new-onset T1D. We will measure pancreas volume both manually and using AI tools. The islets of Langerhans receive a disproportionate share of pancreatic perfusion (10%–20%) given their relative contribution to total pancreatic volume (2%).⁷ Glucose is a potent stimulant for islet cell perfusion, resulting in at least a threefold increase in blood flow.^8,9 We will assess innovative MRI parameters of tissue perfusion dynamics using IVIM and DCE MRI to determine whether hypoperfusion of pancreas parenchyma may be a risk marker for pre-DM or DM. Prior studies have reported a higher prevalence of pancreatic fat fraction in patients with type 2 diabetes (T2D) and obesity.^10-12 We will measure the pancreatic fat signal using MRI, and we expect the fat fraction to be higher in the new-onset DM group compared to the control group. It has been postulated that patients with T1D exhibit exocrine pancreas dysfunction secondary to a lack of insulin and dysregulation of endocrine function.^13,14 Imaging studies have reported a correlation of the T1 signal intensity ratio (SIR) specifically with pancreatic exocrine dysfunction measured by the endoscopic pancreatic function tests.^15,16 Noda et al reported that pancreatic T1 relaxation time is significantly increased with pre-DM individuals and moderately correlates with HbA1c, suggesting that T1 has the potential of being an imaging biomarker for DM.¹⁷ We expect the T1 SIR to be reduced in patients with DM due to the possibility of exocrine dysfunction. Deep learning (DL) is a sub-class of AI wherein layers of neural networks are trained by adjusting parameters; in radiology, it is typically used for detection, segmentation, or classification.¹⁸ We will utilize AI tools, specifically DL algorithms, to automate pancreas and muscle volume measurement and DL algorithms to identify parenchymal changes not visible to the human eye.

MATERIALS AND METHODS

Clinical Imaging and Analysis

Most of the participants in the DREAM study will undergo a cross-sectional examination using standard CT or MRI/MRCP (MR cholangiopancreatography) imaging protocols as part of routine clinical care. Imaging protocols for clinical CT, MRI, and MRCP are developed for the DREAM study; however, we will also utilize any CT or MRI studies performed prior to the participants’ enrollment in the DREAM study. Using the data points listed in Table 1, a site radiologist who has completed protocol-specific training will collect data from clinical studies. Completed case report forms and de-identified cross-sectional studies done for qualifying episode will be uploaded to the central data repository at Penn State University.

TABLE 1.

Parameters Collected From Clinical CT, MRI, and MRCPs

Findings on CT and/or MRI
Pancreas diameter, mm	Head, body and tail
Pancreatic necrosis	Head, body and tail
Venous thrombosis	None Splenic Portal Superior mesenteric vein
Pseudoaneurysm	None Splenic Pancreaticoduodenal Gastroduodenal Other
Fluid collections	Defined in the revised Atlanta classification Acute peripancreatic fluid collection Acute necrotic collection Walled off necrosis Pseudocyst
Location of fluid collections	Lesser sac/perigastric Abdominal retro/intraperitoneal Pelvis/mediastinum
AP Severity Index	a. Pancreatic inflammation. Give score 0, 1, or 2 (0)Normal pancreas (1)Intrinsic pancreatic abnormalities with or without inflammatory changes in peripancreatic fat (2)Pancreatic or peripancreatic fluid collection or peripancreatic fat necrosis
	b. Pancreatic necrosis. Give score 0, 1, 2, 4, or 99 (0)None (1)≤30% (2)>30% and ≤50% (3)>50% (99)Cannot assess (non-contrast study)
	c. Extra-pancreatic findings. Give score 0 or 1 (0)None (1)One or more of the following: pleural effusion, ascites, vascular complications, parenchymal complications, involvement of root of mesentery, or gastrointestinal tract)
Findings on MRI and MRCP
Pancreatic enhancement	Signal intensity on Pre-contrast Arterial Delayed phases
T1 Signal intensity	Pancreas Spleen Paraspinal muscle Liver
Cambridge core	0,1,2,3,4
CBD diameter	mm
CBD stone	Yes/No
Pancreas divisum	Yes/No

AP indicates acute pancreatitis; CBD, common bile duct.

Research MRI Arm (the IMMINENT Study)

A subgroup of up to 250 participants enrolled in longitudinal follow-up in the DREAM study will undergo longitudinal research MRI scans at 3, 12, 24, and 33 months (Fig. 1). Additionally, we will perform a one-time MRI in participants who were not enrolled in longitudinal research MRI; however, develop new-onset DM. MRIs will be performed at 10 of the T1DAPC clinical centers: Indiana University (Indianapolis, Ind), University of Minnesota (Minneapolis, Minn), Cedars Sinai Medical Center (Los Angeles, Calif), University of Illinois at Chicago (Chicago, Ill), AdventHealth System (Orlando, Fla), University of Florida (Gainesville, Fla), Stanford University (Stanford, Calif), Johns Hopkins University (Baltimore, Md), University of Pittsburgh (Pittsburgh, Pa), and the Ohio State University Wexner Medical Center (Columbus, Ohio).

FIGURE 1.

Expected evolution of diabetes status in participants undergoing longitudinal research MRIs in the DREAM study.

Statistical Modelling

Based on published data, we estimate that among 250 participants, 34 participants will develop DM, 43 will develop pre-DM, and 148 will remain euglycemic at 12 months (allowing a 10% loss to follow-up) (Fig. 1). We assume the research MRI participants will exhibit the same rate of loss to follow-up (10% within 12 months and 20% within 24 months). We also assume a 5% annual transitional rate (gray arrows) from euglycemia to DM, a 5% transitional rate from euglycemia to pre-DM, and a 5% transitional rate from pre-DM to DM throughout the study timeline. Given the scarcity of longitudinal imaging studies, our sample size and statistical power calculations are based on Virostko et al,¹⁹ who observed that patients with T1D (not associated with AP) had a pancreas volume 47% smaller than that of euglycemic patients. Specifically, the pancreatic volume index value was 0.69 (standard deviation [SD], 0.37) ml/kg among patients with T1D and 1.2 (SD, 0.35) ml/kg among euglycemic individuals. Based on these index values, we assume the pancreatic volume index will be 0.95 (SD, 0.36) ml/kg among pre-DM patients because pre-DM can be considered a transitional stage between euglycemia and DM. Our application of the Kruskal-Wallis nonparametric analysis of variance shows that there is over 90% statistical power with a two-sided significance level test for comparing the three glycemic groups to the pancreatic volume index. In conjunction with the study protocol, we will incorporate the three-month pancreatic volume index as a regressor in a discrete-time hazards regression model to predict DM incidence. With a sample size of 200 MRI patients at 24 months and a DM incidence of 15%, the precision of the regression model is expressed in terms of a 95% confidence interval of (0.08-0.21) for the DM incidence.

Research MRI Parameters

We will collect several research MRI parameters: three-dimensional (3D) volume using manual segmentation and automated DL algorithms, vascular perfusion using MR perfusion imaging techniques (IVIM and DCE MRI), the pancreatic and liver fat fraction using MRI fat-quantification techniques (Dixon MRI), changes in parenchymal texture using DL algorithms, restriction to diffusion of free water molecules using DWI, T1 mapping and T1-weighted SIR (as a surrogate for pancreatic proteinaceous content), psoas muscle volume (as a surrogate for nutritional status or sarcopenia), and the visceral to subcutaneous fat tissue ratio (as a measure of the pattern of obesity). We will correlate these MRI and AI parameters with the metabolic, genetic, and immunological phenotypes established by the DREAM study.

The Indiana Institute for Biomedical Imaging Sciences will serve as the Core Image Analysis Lab (CIAL) for the research study. The MRI data collection will be performed at the CIAL by experienced MRI research data analysts using dedicated image analysis software. Northwestern University (a satellite site of the Chicago Clinical Center) will serve as the Artificial Intelligence Core Lab (ACL) (Fig. 2). The ACL team will develop DL algorithms to achieve accurate auto segmentation, measurement of gland volume, and subsequent classification of euglycemic, pre-DM, and incident DM cohorts based on glycemic measures collected in the DREAM study.

FIGURE 2.

The flow diagram shows research and clinical MRIs and CTs in the DREAM study.

Research MRI Protocol (the IMMINENT Study)

Magnetic resonance imaging and MRCP will be performed using 3.0T scanners. The generic MRI standard operating procedures shown in Table 2 will be modified and verified for three different hardware vendors (Siemens Medical Solutions USA Inc., Malvern, Pa; GE Healthcare USA, Arlington Heights, Ill; and Phillips Medical Systems, Nashville, Tenn).

TABLE 2.

Standard Operating Procedures for IMMINENT MRI and MRCP

I. Before the MRI evaluation

Verify that patient pass the exclusion criteria. Verify the patient fasted for at least 4 hours prior to the study. Fasting includes any sort of liquids, eg, water, coffee, soda. Any type of liquid will obscure the pancreatic ducts. Assess the patient’s breath-holding capability. If poor, oxygen should be administered.

II. During the MRI evaluation

A. MRI exam guidelines

All imaging must be performed on a 3T scanner. All acquisitions should use a phased-array abdominal coil with a minimum of 8 receive channels. RF transmission will use the built-in body (volume) coil. All patients will receive the same contrast agent: Gadavist® (Gadabutrol, Bayer HealthCare, Whippany, NJ). All patients will receive 0.1 mmol/kg (ml), up to 15 ml (1 vial). The contrast dosage should be the same on all patients regardless of the estimated glomerular filtration rate (eGFR). The contrast will be injected at a 2 mL/sec rate followed by a saline flush of at least 20 mL.

B. Imaging protocol

MRCP

FOV ~30 × 35 cm centered over the pancreas 2D thick slab MRCP, 40 mm thick, breath hold, 6-8 para-coronal projections to best show the pancreatic duct from different angles 3D MRCP: 1 mm respiratory synchronized 3D turbo spin echo sequence, respiratory gated, shallow breathing.

Dixon 2-point fat/water imaging

Axial T1 weighted 3D 2-point Dixon gradient echo sequence capable of generating water-only and fat-only images in addition to in-phase and op-phase images

Pre-contrast T1-weighted images: 3D gradient echo with fat suppression

Acquisition plane: Axial Sequence: 3D T1: LAVA (GE), VIBE (Siemens), or THRIVE (Philips) Fat suppression: ON

T1 mapping pre-contrast

Acquisition plane: axial 3D acquisition dual flip angle spoiled gradient echo sequence Flip angles to be determined by phantom testing

B1 mapping

Acquisition plane: axial Sequence: sites should use their vendor-provided Fast B1 mapping sequences: satTFL (Siemens), Bloch-Siegert (GE), DREAM (Philips) Alternatively, if sites do not have vendor-supplied B1 mapping sequences a 2D SE-EPI may be utilized. Slice thickness/gap: 8 mm/8 mm TA: ~30 s

Intravoxel incoherent motion (IVIM) imaging and diffusion-weighted imaging (DWI)

Sequence: Single-shot echo-planar spin-echo with simultaneous multislice (SMS) imaging Slice thickness/gap: 3 mm/0.6 mm TR/TE/FA: 9700/36 Acquisition plane: axial free breathing Gradient directions: 4 B-values: 0, 50, 100, 150, 400 s/mm2 TA: ~3-4 min

Dynamic contrast enhancement (DCE) imaging

Injected dose weight based at 2 ml/kg Injection rate: 2 ml/s flushed by 20 ml of normal saline Sequence: 3D T1-weighted time-resolved MRA techniques using radial sampling: TRICKS (GE), 3D-TWIST (Siemens), 4D-TRAK (Philips) Slice thickness/gap: 3.5 mm/0 mm Acquisition plane: axial TA: ~2–3 min

T2-weighted images

TSE, or a variant of TSE Acquisition plane: axial Coverage: abdomen only Fat suppression: ON

T2-weighted images

TSE or a variant of TSE Acquisition plane: axial and coronal Coverage: abdomen only Fat suppression: OFF

T1 mapping (post-contrast)

Acquisition plane: Axial Acquisition parameters is same as pre-contrast phase Timing: 12-15 minutes after contrast injection

III. After the MRI evaluation

Confirm that the study data form is completed and will be uploaded together with the images. Studies should be electronically transferred using HIPAA compliant secure FTP (SFTP) protocol to the CIAL for post-processing.

FA indicates flip angle; FOV, field of view; GRAPPA, GeneRalized Auto calibrating Partial Parallel Acquisition; HIPAA, Health Insurance Portability and Accountability Act; LAVA, Liver acquisition with volume acceleration; RF, radiofrequency; SE-EPI, spin-echo echo planar imaging; TA, acquisition time; TE: time to echo; THRIVE, T1W High Resolution Isotropic Volume Examination; TR, repetition time; TRAK, Time-Resolved Angiography using Keyhole; TRICKS, Time-resolved imaging of contrast kinetics; TSE, turbo spin echo; TWIST, Time-resolved angiography With Interleaved Stochastic Trajectories; VIBE, Volumetric interpolated breath-hold examination.

Experienced image analysts will independently measure 3D volume, diameter, IVIM metrics (eg, f, D_fast, D_slow), DCE MRI metrics (e.g., K_trans and K_ve), the T1 SIR, diffusion restriction, the T1-weighted water and fat signals, and visceral and subcutaneous adipose tissue volume. All sites will acquire T1 maps using dual flip angle spoiled gradient echo (SPGR) sequence. It is necessary to standardize the quantitative imaging across different institutions and MR manufacturers using a T1 phantom in order to obtain T1 maps as uniform and accurately as possible. Correction for B1 field inhomogeneity will be incorporated. The T1 SIR of the pancreas will be obtained from unenhanced T1-weighted gradient-echo images with fat suppression by dividing the T1-weighted signal intensity of the pancreas by that of the reference organ (spleen, paraspinal muscle, and liver).¹⁶ The DCE MRI will be acquired using a 3D T1-weighted time-resolved MRA sequence with a single 12° flip angle over a predetermined time course.²⁰ The IVIM sequence will use a free-breathing, single-shot echo-planar imaging design with 4 gradient directions and five b-values: 0, 50, 100, 150, and 400 s/mm².²¹ Our IVIM sequence will use simultaneous multi-slice acquisition simultaneous multi-slice to reduce scan time.²² A propriety post-processing software will be used to model and quantify DCE MRI and IVIM image sets. Dynamic contrast-enhanced MRI processing will begin with generating a T1 map from variable flip angle precontrast T1 data using the classic steady-state, dual-angle linear regression model with corrections for B1 inhomogeneity using two-dimensional double angle spin echo EPI.²³ An arterial input function will be measured by placing a region of interest on the abdominal aorta. The ACL team will simultaneously perform 3D segmentation of the pancreas and then classify study groups based on conventional radiomics/texture analysis and advanced classification with DL algorithms. The Dixon MRI technique can separate the water and fat signals and create fat-only and water-only image sets of the pancreas.²⁴ The pancreatic fat fraction will be calculated by taking the ratio of the signal from fat-only and water-only images. Visceral and subcutaneous spaces will be manually outlined, and volume will be measured using image analysis software. We will apply a fat-segmentation algorithm to automate this process using AI and will measure its success.

Patient Safety and Confidentiality

The MRI scans will be performed under the supervision of an experienced MRI technologist and/or radiology nurse (per practice guidelines at each of the T1DAPC centers). Participants will be screened to confirm that they have none of the contraindications listed in Table 3, have their vital signs documented before and after imaging, and be constantly supervised by a radiology technologist. Per the DREAM study protocol, all serious adverse events that occur within 48 hours following the completion of an MRI scan will be documented and reported. Exposure to high magnetic fields may result in (1) heart rate and rhythm changes (which will be monitored via MR-compatible pulse oximetry or MR-compatible EKG), (2) dizziness, (3) minor nerve stimulation effects, such as muscle twitches and tingling sensations, (4) nausea and vomiting, (5) transient detection of flashes of light, (6) claustrophobia, (7) localized heating of the body, particularly near metal structures, and (8) the displacement of a metallic implant or foreign object in the body by the strong magnetic field. An X-ray of the relevant parts of the body may be performed to determine whether metal objects are present prior to the MRI.

TABLE 3.

Exclusion Criteria for Research MRI as Part of the DREAM Study

a. Unwilling or unable to give a written informed consent.

b. Weight ≥350 lbs.

c. History of moderate or severe allergic reaction to a gadolinium-based contrast agent

d. Pregnancy

e. Claustrophobia that is severe enough to necessitate the use of general anesthesia

f. Known estimated glomerular filtration rate (eGFR) <45 ml/min/1.73m2

g. Severe COPD or other chronic lung disease limiting breath-holding for MRI

h. Moderate or large volume ascites (of any etiology)

i. Hemochromatosis

j. Cystic fibrosis

k. Patients with therapeutic body implants, specifically pacemakers, defibrillators, or other implanted electronic devices (eg, pain pump) that are not MRI-compatible, will be excluded. Patients with an IVC filter, body piercing, neurosurgical clip placement, or shrapnel injury will be evaluated individually.

COPD indicates chronic obstructive pulmonary disease; eGFR, estimated glomerular filtration rate; IVC, inferior vena cava.

Patient confidentially will be ensured through secure processes, such as the DREAM unique ID, that do not expose protected patient data to public disclosure. A secure image-storage server will be established at the CIAL to store the MRI studies. User access is highly restricted because only the secure file transfer protocol (SFTP) server will operate in a jailed environment within the Indiana University cluster of servers. The study server is HIPAA (Health Insurance Portability and Accountability Act) compliant and will be situated behind network firewalls. The SFTP server will be set up to restrict access to fixed internet provider numbers; therefore, only the designated locations will have access. In addition, the study-site users will use the RSA-4096-bit key to authenticate to the SFTP server, ensuring that user credentials cannot be intercepted. Once connected, centers user will upload their MR images, which will be automatically sent to an anonymizer pipeline. This pipeline will eliminate any secondary capture images, reports, etc. The de-identified images will be sent to the temporary storage server. Access to the de-identified data in this server will be limited to selected individuals. At the end of the image analysis, research MRI scans will be uploaded to the T1DAPC permanent image repository.

Artificial Intelligence Analysis

In this study, we will use existing AI, specifically DL, tools and develop new ones to perform several tasks: measurement of 3D volume, pancreas texture analysis, and assessment of sarcopenia by measuring psoas muscle volume. The ACL team will modify their deep-capsule-based segmentation networks (SegCaps) and apply them to the MR sequences. For volumetric (3D) segmentation, SegCaps’ internal mechanism will be used. It will be compared with other pancreas-segmentation baselines that utilize conventional slice-by-slice image segmentation (ie, pseudo-3D), such as U-Net and its variants. We will improve the segmentation algorithms with newly developed theoretical advances (such as attention mechanisms by Transformers) to mimic the shape and appearance differences of the pancreas better than before. We expect that the proprietary software developed by the Northwestern University team will segment the pancreas accurately and enable a quantitative analysis of the pancreas. In addition, conventional and DL-based radiomics/texture analysis and other quantitative MRI marker evaluations will be performed for comparisons and benchmarking.

DISCUSSION

The DREAM study presents a unique opportunity to prospectively monitor MRI biomarkers in the pancreas and relate them directly to imaging, metabolic and immunological markers of DM. This methodology paper describes the design and rationale for using cross-sectional clinical imaging combined with longitudinal research MRI scans to predict DM following AP. Several preliminary studies have suggested that MRI of the pancreas can detect changes associated with the development of DM. In addition to providing anatomical information, MRI offers a rich toolbox of imaging approaches that are sensitive to a large variety of physiologic and metabolic phenomena such as volume, fat and protein content, fibrosis, and blood flow. Based on this evidence, we propose that novel MRI techniques can be used to understand pathophysiology within the pancreatic parenchyma and can offer methods for defining the natural history of DM after AP.²⁵ These data will be needed to understand the natural history of morphological, functional, and pathophysiological changes that occur before the development of pre-DM and DM. We will correlate several MR imaging parameters (eg, perfusion metrics, 3D volume, imaging severity of AP, T1 signal, fat fraction, texture differences, etc.) with participant demographics, etiology of AP, clinical severity of AP, and disease outcome (euglycemic, pre-DM vs DM). In addition, we will stratify imaging parameters in different phenotypes, including but not limited to the type of DM (T1D, T2D, or type 3c DM), laboratory (fasting blood glucose, HbA1c), immunologic phenotypes (beta-cell autoantibodies), EPD, alterations in the intestinal microbiome, dietary patterns or hormonal status (insulin secretion, glucagon, and pancreatic polypeptide). This analysis will be performed using statistical analysis and/or DL algorithms where a wide variety of clinical data can be analyzed. Eventually, we plan to develop a quantitative composite risk score based on useful MRI parameters. This noninvasive and practical imaging biomarker may enable monitoring of the evolution of metabolic dysfunction in clinical practice or clinical trials. These approaches may result in noninvasive imaging biomarkers to identify specific properties of the pancreas associated with DM development. The standardization benefit of quantitative MRI techniques allows more accurate comparison across different platforms, therefore permitting a more useful interpretation.

ABBREVIATIONS

DREAM

Diabetes Related to Acute Pancreatitis and Its Mechanisms

IMMINENT

Imaging Morphology of Pancreas in Diabetic Patients Following Acute Pancreatitis

MRI

Magnetic Resonance Imaging

DCE-MRI

Dynamic contrast-enhanced MRI

IVIM

Intravoxel incoherent motion

DWI

Diffusion-weighted imaging

MRCP

MR cholangiopancreatography

AP

Acute Pancreatitis

Pre-DM

Prediabetes

DM

Diabetes Mellitus

T1D

Type 1 diabetes mellitus

T2D

Type 2 diabetes mellitus

CIAL

Core Image Analysis Lab

ACL

Artificial Intelligence Core Lab

AI

Artificial Intelligence

T1DAPC

Type 1 Diabetes in Acute Pancreatitis Consortium

SIR

Signal Intensity Ratio

EPD

Exocrine Pancreas Dysfunction

CBD

common bile duct

eGFR

estimated glomerular filtration rate

DL

Deep learning

SegCaps

deep capsule-based segmentation networks