Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 May 26.
Published before final editing as: Clin Anat. 2026 Jan 20:10.1002/ca.70077. doi: 10.1002/ca.70077

Lifespan Pancreas Morphology for Control Versus Type 2 Diabetes Using AI on Largescale Clinical Imaging

Lucas W Remedios 1, Chloe Cho 2, Trent M Schwartz 3, Dingjie Su 1, Gaurav Rudravaram 3, Chenyu Gao 3, Aravind R Krishnan 3, Adam M Saunders 3, Michael E Kim 1, Shunxing Bao 3, Thomas A Lasko 1,4, Alvin C Powers 5,6,7, Bennett A Landman 1,2,3, John Virostko 8,9,10,11
PMCID: PMC13200214  NIHMSID: NIHMS2173214  PMID: 41556492

Abstract

Understanding how pancreas size and shape change with normal aging is critical for establishing a baseline to detect deviations in type 2 diabetes and other pancreatic disease. We measure pancreas size and shape using morphological measurements from early development through aging (ages 0–90). Our goals are to (1) identify reliable clinical imaging modalities for artificial intelligence (AI) based pancreas measurement, (2) establish normative morphological aging trends, and (3) detect potential deviations in type 2 diabetes. We analyzed a clinically acquired dataset of 2533 patients imaged with abdominal computed tomography (CT) or magnetic resonance imaging (MRI). The patients did not have cancer, pancreas pathology, sepsis, or trauma. We resampled the scans to 3 mm isotropic resolution, segmented the pancreas using automated methods, and extracted 13 morphological pancreas features across the lifespan. First, we assessed pancreas volume trajectories in 1858 control patients across contrast CT, non-contrast CT, and MRI to determine which modalities provide consistent lifespan trends. Second, we characterized distributions of normative morphological patterns stratified by age group and sex. Third, we used covariate-adjusted generative additive models for location, scale, and shape (GAMLSS) regression to model pancreas morphology trends in 1350 patients matched for age, sex, and type 2 diabetes status to identify any deviations from normative aging associated with type 2 diabetes. We selected CT for the main analyses of this study, since the MRI appeared to yield different pancreas measurements than CT using our AI-based method on this dataset of clinically acquired scans. When adjusting for confounders, the aging trends for 10 of 13 morphological features were significantly different between patients with type 2 diabetes and non-diabetic controls (p < 0.05 after multiple comparisons corrections). Additionally, we characterized normative morphological aging trends of the pancreas across 13 morphological measurements. We provide lifespan trends demonstrating that the size and shape of the pancreas are altered in type 2 diabetes using 675 control patients and 675 diabetes patients. Moreover, our findings reinforce that the pancreas is smaller in type 2 diabetes. Additionally, we contribute a reference of lifespan pancreas morphology from a large cohort of non-diabetic control patients in a clinical setting.

Keywords: aging, CT, MRI, multimodal, pancreas, shape, volume

1 ∣. Introduction

Substantial efforts have been made to measure the morphology of the structural aging of organs, such as the brain, which helps distinguish between atrophy caused by normal aging and atrophy caused by disease (Statsenko et al. 2023). Defining an organ’s normative morphological aging allows investigation into how disease alters this trajectory, offering insight into the disease process. As with the brain, the pancreas also undergoes changes with age (Löhr et al. 2018) (Figure 1).

FIGURE 1 ∣.

FIGURE 1 ∣

Open in a new tab

The pancreas undergoes structural changes with age, including atrophy and fat infiltration. While population-level pancreas volume and fat content have been examined across the aging process (Saisho et al. 2007), there remains a knowledge gap in understanding age-related changes in the pancreas across a broader set of morphological measurements. Moreover, type 2 diabetes may cause changes in pancreas morphology that differ from normal aging. The two scans on the left illustrate age-related appearance differences in non-diabetic patients but are not from the same patient. The rightmost scan shows the pancreas from an elderly patient with type 2 diabetes. Understanding pancreas variation in normal aging is critical for understanding differences in type 2 diabetes. Any potential differences in the aging trends of the pancreas in type 2 diabetes may not necessarily be linear or smooth.

As the pancreas ages, it changes in size, shape, fat content, and duct structure, along with increasing fibrosis (Möller et al. 2023; Hastier et al. 1998; Saini et al. 2015; Glaser and Stienecker 2000; Kreel and Sandin 1973; Gupta et al. 2017). Separately from aging, pancreas morphology is influenced by sex and body composition (Dogan et al. 2021; Wang et al. 2021). Additionally, pancreatic morphological alterations have been extensively studied and reported in type 2 diabetes.

Pancreas volume is a critical measure of pancreas morphology. Nearly 20 years ago, Saisho et al. (2007) measured pancreas volume across the lifespan, where the pancreas was segmented (outlined by hand) from computed tomography (CT). Today, artificial intelligence (AI) enables faster, fully-automated segmentation of the pancreas (Dogan et al. 2021). Pancreas volume declines with age and is partially replaced by ectopic fat (Saisho et al. 2007; Löhr et al. 2018; Chantarojanasiri et al. 2015). In type 2 diabetes, the pancreas has been found to be smaller than control subjects, and has a serrated edge (Desouza et al. 2018; Macauley et al. 2015). In Saisho et al. (2007), the pancreas was determined to be smaller in subjects with type 2 diabetes in a large dataset matched for age, sex, and body mass index (BMI)—this finding aligned with previous studies (Fonseca et al. 1985; Alzaid et al. 1993; Gilbeau et al. 1992; Klöppel et al. 1985; Migdalis et al. 1991), but also differed from other studies that found no volume difference in type 2 diabetes from control subjects (Silva et al. 1993; Garcia et al. 2017). However, it is worth noting that the Silva et al. (1993) study was from ultrasound in the early 1990’s when precision was lower.

Beyond volume, deep characterization of pancreas morphology may lead to insights on disease states. The diameter of the pancreas has been identified as an important feature of pancreatic health (Desouza et al. 2018). Additionally, pancreas shape features have been extracted on magnetic resonance imaging (MRI) to study nearly 4000 subjects (Bagur et al. 2020).

In this work, we measure the pancreas from a large clinical dataset via AI-driven pancreas segmentation (Figure 2). We comment on the consistency of clinical CT and MRI for volume measurements and present lifespan trends on normative aging with 13 morphological features of the pancreas. Further, we test for differences in structural aging between control and type 2 diabetes in a dataset matched on sex and age spanning ages 20 to 90. Rather than correcting for body size through division, we model aging trends while accounting for the covariates (age, sex, weight, and diabetes status).

FIGURE 2 ∣.

FIGURE 2 ∣

Open in a new tab

We leverage over 2500 clinical abdominal scans (CT or MRI) from control and type 2 diabetes patients. Using the AI tool TotalSegmentator (Wasserthal et al. 2023; Akinci D’Antonoli et al. 2025), we automatically segment the pancreas. We then use PyRadiomics (van Griethuysen et al. 2017) to extract 13 morphological measurements across the lifespan to assess whether pancreas aging differs in type 2 diabetes. We control for age, sex, and weight effects through both matching and modeling.

2 ∣. Methods

2.1 ∣. Data

Our dataset was retrieved in deidentified form from site and IRB redacted for anonymization. We extracted features from an initial pool of 32,894 clinically acquired medical images. Because these data were obtained for routine clinical care rather than under a standardized research protocol, only a subset met our stringent quality assurance and inclusion criteria detailed below. After full processing and cohort selection, there were 2533 patients, each with one abdominal scan (CT or MRI), and each with a single diabetes label, either control or type 2 diabetes. For compatibility with downstream processing, the images were converted from DICOM to NIfTI format via dcm2niix (Li et al. 2016).

2.2 ∣. AI Segmentation & Feature Extraction

The scans were converted into LAS orientation, cropped between the L5 and T7 vertebrae (TotalSegmentator (Wasserthal et al. 2022) vertebrae segmentation), and resampled to 3 mm isotropic resolution. The pancreas and other abdominal organs were segmented with TotalSegmentator (Wasserthal et al. 2022; D’Antonoli et al. 2024) version 2.8, using the CT model for CT scans, and the MRI model for MRI scans. Thirteen morphological features were extracted from the binary NIfTI pancreas segmentations using PyRadiomics (van Griethuysen et al. 2017) version 3.1. The features were: volume, surface area, surface area to volume ratio, elongation, flatness, sphericity, major axis length, minor axis length, least axis length, maximum 3D diameter, maximum 2D diameter column, maximum 2D diameter row, and maximum 2D diameter slice.

2.3 ∣. Quality Control of Medical Images & Segmentations

To ensure data quality, all images and organ segmentations were manually inspected with a high-throughput visualization tool (Kim et al. 2024). In some cases, the conversion from DICOM to NIfTI format failed and corrupted image metadata. In these DICOM to NIfTI failure cases, the images were corrupted and exhibited geometric warping to varying degrees across the axial, sagittal, and coronal axes. The resulting warping produced distorted representations of pancreatic and abdominal anatomy that did not reflect true anatomical structure and would confound quantitative morphometric measurements if retained. Additionally, undesirable cases contained organ segmentations with more than one contiguous volume or that touched the edge of the image volume, indicating the organ extended beyond the field-of-view. These scans were excluded, with the warping/corruption identified via out-of-distribution examples of the ratio of the faces on a bounding box of the liver segmentation.

A subsequent manual inspection of the images and segmentations revealed that scans corrupted by geometric warping persisted in the dataset following these initial exclusion steps. In imaging sessions containing multiple acquisitions, such corruption affected only a subset of scans, necessitating a procedure to identify the uncorrupted acquisition within each session. To support this technical quality-control step, variability in organ volumes across repeated scans within the same session was evaluated across multiple abdominal organs (pancreas, liver, spleen, left kidney, and right kidney). Smooth polynomial reference trends across age were used solely as a technical envelope to identify implausible volume distortions consistent with the geometric warping artifacts, rather than as a model of biological variation. Within each session, scans were ranked based on their deviation from these reference trends with equal weighting across the five organs, and the scan least affected by such distortion was retained, while scans exhibiting warping artifacts/corruption were excluded. This approach involved a tradeoff between removing corrupted data and preserving biological variability; however, it was applied exclusively to address imaging artifacts and not to select scans based on biological characteristics or downstream analysis outcomes. All retained scans were subsequently verified by manual inspection, and all warped scans were excluded from the dataset.

2.4 ∣. Diabetes Label Assignment

To assign diabetes labels to the patients, we used their health records up to 1 year after the selected scan. International classification of disease (ICD) codes and derived PhecodeX (Shuey et al. 2023) were used to determine control patients as those without any diabetes events, and patients with type 2 diabetes as those with at least one type 2 diabetes event and no type 1 diabetes events. Where available, A1C measurements were used to validate that the control patients did not have diabetes. Using diagnosis information from ICD and PhecodeX, as well as procedure information from their current procedural terminology (CPT) and procedure wide association (ProWAS) codes, we excluded patients with cancer, pancreas pathology, and sepsis (Chaganti et al. 2017; Kerley et al. 2022). Pancreas pathology was identified by filtering ICD and PhecodeX records for the word segment “panc.” We also excluded imaging associated with trauma events to avoid scans with acute structural and functional alterations (e.g., hemorrhage, edema).

2.5 ∣. Automatic Contrast Phase Labeling of CT

CT scans were automatically labeled according to contrast phase using TotalSegmentator’s contrast phase prediction tool (Wasserthal et al. 2022). These labels were inspected and determined to be accurate for binary contrast labeling (contrast CT or non-contrast CT). The final binary phase classification was manually assured for each CT through visual inspection of the scans using AutoQA (Kim et al. 2024).

2.6 ∣. Pancreas Measurement Consistency Across CT & MRI

Before analyzing anatomical trends, we needed to select which imaging modalities (CT or MRI) in the clinical dataset consistently measured the pancreas. We chose to use pancreas volume, pancreas volume index (volume divided by patient weight), and BMI of healthy controls to assess this. Based on observation of lifespan polynomials fit with 95% confidence intervals, we selected the subset of modalities to use in subsequent analyses.

2.7 ∣. Lifespan Pancreas Morphology in the Clinic by Sex

On our selected modalities, we created a lifespan reference of normative pancreas morphology by creating boxplots to visualize the distributions of each uncorrected pancreas morphological feature by sex and age group as in Saisho et al. (2007). These data were separated by sex but were not matched.

2.8 ∣. Modeling Pancreas Aging With and Without Diabetes

We modeled each pancreas feature across the adult lifespan (ages 20–90) using a single generalized additive model for location, scale, and shape (GAMLSS (Stasinopoulos and Rigby 2007)), jointly incorporating both male and female, as well as individuals with and without type 2 diabetes. The location (μ) was modeled as:

μ=β0+f(age)+β1diabetes+β2sex+β3weight (1)

Here β0 is the intercept, and β1, β2, and β3 are coefficients for type 2 diabetes status, sex, and weight respectively. The term (age) represents a smooth, nonlinear function of age, implemented using penalized B-splines. We selected this GAMLSS model to flexibly capture the nonlinear nature of biological aging. In contrast, diabetes, sex, and weight were modeled linearly under the assumption that they introduce global shifts. The same covariates were included in the model for the scale (σ), and we used the Box-Cox Cole and Green (BCCG) distribution to handle skewness in the feature distributions. Although GAMLSS does not include an explicit residual term like in traditional linear models, the progression-related variability (i.e., error) is captured through the distribution parameters. The GAMLSS approach enables learning distributions, rather than just the mean, as a function of the input variables.

Although one model was fit per pancreas feature, we visualized the fitted curves separately for male and female to highlight sex-specific differences. For each sex, weight was fixed to the average weight across all patients of that sex (both type 2 diabetes and control), and we conditioned the visualization on type 2 diabetes status to isolate its impact on pancreas aging. Because we modeled type 2 diabetes diagnosis as a linear relationship, we obtained a single p-value for the differences between the type 2 diabetes and control groups. These p-values were corrected using the Benjamini– Hochberg False Discovery Rate (FDR) to account for the 13 features tested. After FDR correction, p < 0.05 was considered significant.

2.9 ∣. Use of Generative AI in Manuscript

Generative AI (ChatGPT4o) was employed to assist with drafting and content refinement throughout this research. All core intellectual content and insights originated from the authors’ independent scholarly work. All AI-generated material passed through careful author review and revision to ensure alignment with the study’s objectives and to maintain academic rigor.

3 ∣. Results

3.1 ∣. Comparison of Clinical Imaging Modality for Measuring the Pancreas

In Figure 3, we observed for agreement between contrast CT, non-contrast CT, and MRI for measuring the pancreas via lifespan trends (polynomials). We selected pancreas volume as the main feature, since population-level reference ranges are provided by Saisho et al. (2007). We used an unmatched subset of the dataset consisting of all 1858 control patients. We observed that MRI tended to yield smaller pancreas measurements than CT. To account for differences in weight distributions, we also assessed the pancreas volume index (pancreas volume divided by patient weight), where the pattern was still observed. We additionally checked the distribution of BMI across modalities, which was comparable. Based on this observed incongruency between image modalities, we cautiously excluded automated pancreas segmentation derived from MRI from our subsequent analyses and instead focused on the larger group of CT images.

FIGURE 3 ∣.

FIGURE 3 ∣

Open in a new tab

We investigate measurement consistency between CT and MRI using lifespan polynomials. With our automatic AI approach, MRI appears to measure a smaller pancreas volume than CT for most age groups (A). The observable difference between the MRI and CT trends becomes more evident when correcting for body size by dividing pancreas volume by patient weight (B). BMI is comparable across the modalities, which further implies that observed reduced pancreas measurements in MRI are likely not caused by smaller body size (C). These trends are from non-diabetic control patients. Given the observed discrepancy in automated pancreas volume measures between MRI and CT, we cautiously exclude the smaller cohort of MRI measurements from further analyses and proceed with CT measurements.

3.2 ∣. Lifespan Pancreas Morphology Shows Aging Trends

In Figure 4, we present uncorrected lifespan morphology of the pancreas in an unmatched dataset of 1775 nondiabetic control patients with CT imaging of the pancreas. Our results reproduce the population volume trends from Saisho et al. (2007) Across size metrics (axis lengths and diameters), the female pancreas is generally smaller in adulthood.

FIGURE 4 ∣.

FIGURE 4 ∣

Open in a new tab

From a population of 1775 non-diabetic control patients with CT scans, we illustrate how pancreas size and shape change with age and sex across 13 morphological measurements. We reproduce findings from Saisho et al. on pancreas volume (Saisho et al. 2007). These distributions visualize population spread of pancreas measurements across age groups.

3.3 ∣. Matched Dataset for Assessing Control Versus Type 2 Diabetes

In Figure 5, we show the distribution of matched patients by age, sex, and diabetes status. These 1350 patients allow a large sample size for measuring potential adulthood pancreas aging differences in type 2 diabetes, with most of the data being available from age 40 to 70.

FIGURE 5 ∣.

FIGURE 5 ∣

Open in a new tab

We used a matched subset of the data with 1350 patients to mitigate the confounding effects of sex and age before assessing whether pancreas morphology differs in type 2 diabetes. Body size was not matched but was addressed later through modeling. Each cell provides the number of patients, colored by the color bar, in each diabetes/sex and age group.

3.4 ∣. Shift in Diabetes

In Figure 6, we show that when accounting for covariates, 10 of 13 morphological measurements of the pancreas were significantly different (p < 0.05 after multiple comparisons corrections on the linear type 2 diabetes parameter). On the statistically significant features, there were small shifts between the type 2 diabetes and control curves, indicating that type 2 diabetes is associated with measurable changes in pancreas morphology throughout adulthood. These differences persisted even after both matching and modeling the data to account for age, sex, and body weight using GAMLSS. Significant differences were observed across most of the anatomical measurements, including pancreas volume, surface area, surface-to-volume ratio, elongation, flatness, sphericity, and multiple axis and diameter-based measurements. We also observed sex-specific trends, with shifts between male and female trajectories. The findings in Figure 6 support that the type 2 diabetes pancreas is smaller than control and has altered morphology throughout aging in adulthood, however there is a large amount of overlap in the distributions between type 2 diabetes and control. In Table 1, we complement Figure 6 by reporting decade-level reference values for pancreatic measurements derived from the same GAMLSS model used in Figure 6. For each measurement, the GAMLSS model estimated the median (50th percentile) at each integer age from 20 to 90 years. The decade-level values in Table 1 were obtained by taking the median of these age-specific model-estimated medians within each decade. In Figure 7, we present complementary numerical details for the type 2 diabetes coefficient from the μ of the GAMLSS model.

FIGURE 6 ∣.

FIGURE 6 ∣

Open in a new tab

In type 2 diabetes, the pancreas is smaller and has an altered shape with our GAMLSS regression that models diabetes diagnosis with a linear term. Across these 13 morphological measurements, 10 of the aging trends were significantly different in diabetes, with p < 0.05 after multiple comparisons correction denoted with *. In these plots, we show curves that represent the 50th percentile (median) of the GAMLSS model-learned distributions, with the range from the 5th to 95th centiles shaded, holding other covariates constant. While we detect statistical significance via the linear type 2 diabetes parameter across most of the metrics, the distributions from the patients with type 2 diabetes are similar to the nondiabetic controls.

TABLE 1 ∣.

Complementary to Figure 6, we present decade-level pancreatic reference values derived from the same GAMLSS model.

Model-estimated
measurements		 Sex Diabetes
status		  20–29 30–39 40–49 50–59 60–69 70–79 80–89
Volume (mL) M Control 81.898 82.857 83.722 82.344 77.391 71.362 66.338
M T2DM 79.975 80.934 81.798 80.420 75.467 69.438 64.415
F Control 69.672 70.631 71.496 70.117 65.165 59.136 54.112
F T2DM 67.748 68.708 69.572 68.194 63.241 57.212 52.189
Surface area (cm2) M Control 170.119 170.479 173.003 173.143 168.472 161.443 154.516
M T2DM 166.800 167.160 169.685 169.824 165.154 158.125 151.197
F Control 149.305 149.665 152.190 152.329 147.659 140.629 133.702
F T2DM 145.986 146.347 148.871 149.011 144.340 137.311 130.384
Surface area to volume ratio (mm2/mm3) M Control 0.207 0.207 0.208 0.213 0.222 0.232 0.243
M T2DM 0.208 0.208 0.209 0.214 0.223 0.234 0.245
F Control 0.216 0.217 0.218 0.222 0.231 0.242 0.253
F T2DM 0.218 0.218 0.219 0.224 0.233 0.243 0.254
Elongation (unitless) M Control 0.349 0.362 0.372 0.379 0.378 0.374 0.368
M T2DM 0.359 0.371 0.382 0.389 0.388 0.383 0.377
F Control 0.341 0.353 0.364 0.371 0.370 0.365 0.359
F T2DM 0.350 0.363 0.373 0.380 0.379 0.375 0.369
Flatness (unitless) M Control 0.201 0.199 0.197 0.195 0.193 0.191 0.189
M T2DM 0.209 0.207 0.205 0.203 0.200 0.198 0.197
F Control 0.217 0.215 0.214 0.211 0.209 0.207 0.205
F T2DM 0.225 0.223 0.221 0.219 0.216 0.214 0.213
Sphericity (unitless) M Control 0.543 0.537 0.531 0.524 0.516 0.509 0.501
M T2DM 0.544 0.538 0.532 0.525 0.517 0.509 0.502
F Control 0.550 0.545 0.539 0.532 0.524 0.516 0.509
F T2DM 0.551 0.545 0.539 0.532 0.524 0.517 0.509
Major axis length (mm) M Control 159.243 160.349 161.456 162.563 163.668 164.772 165.876
M T2DM 155.901 157.008 158.115 159.221 160.327 161.431 162.535
F Control 143.469 144.576 145.682 146.789 147.894 148.999 150.102
F T2DM 140.128 141.235 142.341 143.448 144.553 145.657 146.761
Minor axis length (mm) M Control 55.779 57.998 60.072 61.676 61.944 61.250 60.132
M T2DM 56.118 58.337 60.411 62.016 62.283 61.589 60.471
F Control 48.416 50.636 52.709 54.314 54.581 53.887 52.770
F T2DM 48.756 50.975 53.048 54.653 54.920 54.226 53.109
Least axis length (mm) M Control 32.114 31.999 31.884 31.768 31.653 31.538 31.422
M T2DM 32.619 32.504 32.388 32.273 32.158 32.042 31.927
F Control 31.258 31.143 31.027 30.912 30.796 30.681 30.566
F T2DM 31.763 31.647 31.532 31.417 31.301 31.186 31.071
Maximum 3D diameter (mm) M Control 148.115 149.556 150.995 152.305 153.403 154.203 154.836
M T2DM 145.716 147.157 148.596 149.906 151.004 151.804 152.437
F Control 135.953 137.394 138.833 140.143 141.241 142.040 142.673
F T2DM 133.554 134.995 136.434 137.744 138.842 139.642 140.274
Maximum 2 D diameter column (mm) M Control 110.291 111.864 113.643 115.292 116.114 115.982 115.574
M T2DM 112.570 114.143 115.922 117.571 118.394 118.261 117.853
F Control 99.717 101.290 103.069 104.718 105.541 105.408 105.000
F T2DM 101.996 103.570 105.348 106.997 107.820 107.688 107.279
Maximum 2D diameter row (mm) M Control 63.382 64.881 66.077 66.749 66.532 65.532 64.306
M T2DM 64.642 66.140 67.336 68.008 67.792 66.791 65.565
F Control 59.087 60.585 61.781 62.453 62.236 61.236 60.010
F T2DM 60.346 61.845 63.041 63.713 63.496 62.495 61.270
Maximum 2D diameter slice (mm) M Control 122.471 123.683 125.076 125.902 125.729 124.765 123.758
M T2DM 117.515 118.727 120.120 120.946 120.774 119.810 118.803
F Control 105.678 106.890 108.283 109.109 108.937 107.973 106.966
F T2DM 100.722 101.934 103.328 104.154 103.982 103.019 102.014
Open in a new tab

Note: Values represent the median obtained by aggregating age-specific model-estimated medians (50th percentiles) within each decade.

FIGURE 7 ∣.

FIGURE 7 ∣

Open in a new tab

We complement the curves from Figure 6 with the numerical results for the type 2 diabetes linear coefficient from the μ of the learned distributions from the GAMLSS model. Statistical significance after FDR correction is denoted in green.

4 ∣. Discussion

In this study, we characterized age-related changes in pancreas size and shape using a large, clinically acquired dataset and AI-based segmentation across CT and MRI modalities. We found that pancreas morphology ages differently in type 2 diabetes across numerous measures of size and shape. Our modality comparison led to observed differences between MRI and CT in AI-based pancreas measurement, emphasizing the need for modality-specific baselines.

Measuring the pancreas from medical imaging is commonly performed on both CT and MRI in research settings. In clinical populations, CT is more prevalent; however, MRI is commonly performed in research studies to avoid radiation exposure (Virostko et al. 2021). In datasets of clinically acquired abdominal CT scans, images reflect real-world variability in image quality and acquisition protocols, which may influence pancreas appearance and measurement. Several previous studies have inspected the pancreas from CT and/or MRI (Syed et al. 2012; Caglar et al. 2012; Kipp et al. 2019; Zhou et al. 2025; Sato et al. 2012; Le Goallec et al. 2022; Remedios et al. 2025).

We observed that, within this clinically acquired dataset, AI-based pancreas volume measurements derived from MRI were systematically smaller than those derived from contrast and non-contrast CT. However, our AI-based measurements on CT reproduced previously reported lifespan trends from manual pancreas volumetry (e.g., Saisho et al. 2007), suggesting that the observed discrepancy in pancreas size between CT and MRI was specific to the application of our automated segmentation approach across modalities rather than to biological differences. Consistent with this observation, manual quality assurance identified a higher frequency of segmentation failures on MRI than on CT. This likely reflects challenges associated with the generalization of pretrained AI-based models on heterogeneous routine clinical MRI acquisitions without dataset-specific adaptation or harmonization.

Prior CT- and MRI-based studies have generally reported reduced pancreatic volume in individuals with type 2 diabetes. Saisho et al. (2007), for example, matched on sex, age, and BMI and found that the pancreas of people with type 2 diabetes was smaller than control. Our findings agree with the evidence pointing toward a smaller pancreas in type 2 diabetes. Moreover, we demonstrate that the shape of the pancreas is also altered in type 2 diabetes across several morphological measurements. Importantly, recent work has demonstrated that substantial weight loss in individuals with type 2 diabetes can return pancreas size and surface irregularity to normal (Al-Mrabeh et al. 2020). While an additional investigation of fat infiltration across the lifespan could prove insightful, the vast majority of our CT had contrast, which renders fat estimation from Hounsfield units inaccurate.

To compare subjects with different body sizes, studies often divide pancreas volume by body surface area (BSA), weight, or BMI—an approach sometimes used to assess differences between individuals with and without diabetes (Goda et al. 2001; Al-Mrabeh et al. 2016; Cai et al. 2025; Wright et al. 2022; Pollé et al. 2023; Vesterhus et al. 2008; Almeida et al. 2018; Fortson et al. 2024; Kawaji et al. 2021; Virostko et al. 2021). This approach is supported by demonstrations that there are correlations between pancreas volume and BSA, weight, and BMI (Macauley et al. 2015; Caglar et al. 2014; McCleary et al. 2020; Sequeira et al. 2022). In our comparison between patients with type 2 diabetes and non-diabetic controls, we accounted for body size variability by including weight as a covariate.

This work is limited in that we consider contrast in CT as a binary flag. In reality, contrast cycles through phases which impact the way the pancreas appears on imaging. Variability in contrast phase may influence pancreas segmentation via TotalSegmentator and may influence resulting measurements of the pancreas.

To advance our understanding of pancreas aging, future work should focus on quantifying changes in pancreatic size and shape using higher-resolution medical imaging to enable finer structural analysis. In parallel, evaluating how anatomical aging patterns vary across clinical sites and populations will help establish robust baselines. In this study, we focused on CT scans, since automated pancreas measurements differed between CT and MRI in this dataset and the available CT cohort was substantially larger. MRI-based methods can likewise provide precise anatomical information when applied under standardized acquisition protocols in controlled research settings; however, extending such AI-based analyses to large-scale collections of routine clinical MRI remains an active area of methodological development due to variability in acquisition parameters.

5 ∣. Conclusions

We find that in patients with type 2 diabetes, the pancreas is significantly smaller and different in shape than non-diabetic controls via a linear parameter for type 2 diabetes in the GAMLSS model. We provide reference trends for how control and type 2 diabetes pancreas morphology changes across the lifespan, as imaged in the hospital. The observed differences in AI-based pancreas measurements between clinically acquired CT and clinically acquired MRI in this large-scale dataset highlight that for cross-modality studies, it is essential to use robust image harmonization to ensure reliable pancreas measurements across imaging types. Without image harmonization, findings in one imaging modality may be difficult to interpret in the context of another.

Acknowledgments

This work was supported by Integrated Training in Engineering and Diabetes, grant number T32 DK101003 and MSTP T32: NIH NIGMS T32GM007347. This work used REDCap and VCTRS resources, which are supported by grant UL1 TR000445 from National Center for Advancing Translational Sciences, National Institutes of Health (NIH). This work received support from the Vanderbilt Diabetes Research and Training Center (DK020593), Division of Diabetes, Endocrinology, and Metabolic Diseases, and the Vanderbilt University Institute of Imaging Science Center for Human Imaging (1 S10OD021771 01). This work was supported by the Alzheimer’s Disease Sequencing Project Phenotype Harmonization Consortium (ADSP-PHC) that is funded by NIA (U24 AG074855, U01 AG068057 and R01 AG059716). This work was supported by NSF career 1452485 and NSF 2040462. This research was funded by the National Cancer Institute (NCI) grant R01 CA253923-04, R01 CA 253923-04S1. This work was conducted in part using the resources of the Advanced Computing Center for Research and Education at Vanderbilt University, Nashville, TN. The Vanderbilt Institute for Clinical and Translational Research (VICTR) is funded by the National Center for Advancing Translational Sciences (NCATS) Clinical Translational Science Award (CTSA) Program, Award Number 5UL1TR002243-03. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. This work was supported by DoD grant HT94252410563. We extend gratitude to NVIDIA for their support by means of the NVIDIA hardware grant. Financial support was graciously provided by the National Institutes of Health (DK129979 and HD115565) and Breakthrough T1D (formerly JDRF) (1-INO-2023-1340-A-N). We gratefully acknowledge philanthropic support from Thomas J. and Karen K. Gentry. The study sponsors were not involved in the design of the study and did not impose any restrictions regarding the publication of the report. We have used AI as a tool in the creation of this content; however, the foundational ideas, underlying concepts, and original gist stem directly from the personal insights, creativity, and intellectual effort of the author(s). The use of generative AI serves to enhance and support the author’s original contributions by assisting in the ideation, drafting, and refinement processes. All AI-assisted content has been carefully reviewed, edited, and approved by the author(s) to ensure it aligns with the intended message, values, and creativity of the work.

Funding

This work was supported by the National Institutes of Health (T32 DK101003, NIGMS T32GM007347, UL1 TR000445, DK020593, 1 S10OD021771 01, U24 AG074855, U01 AG068057, R01 AG059716, R01 CA253923-04, R01 CA 253923-04S1), the National Science Foundation (1452485, 2040462), the National Institutes of Health (5UL1TR002243-03, DK129979, HD115565), the U.S. Department of Defense (HT94252410563), the Breakthrough T1D (1-INO-2023-1340-A-N), and the Philanthropic support from Thomas J. and Karen K. Gentry.

Footnotes

Disclosure

Our dataset was retrieved in deidentified form from Vanderbilt University Medical Center under IRB #241494.

Data Availability Statement

Research data are not shared.

References

  1. Akinci D’Antonoli T, Berger LK, Indrakanti AK, et al. 2025. “TotalSegmentator MRI: Robust Sequence-Independent Segmentation of Multiple Anatomic Structures in MRI.” Radiology 314, no. 2: e241613. [DOI] [PubMed] [Google Scholar]
  2. Almeida RR, Lo GC, Patino M, Bizzo B, Canellas R, and Sahani DV. 2018. “Advances in Pancreatic CT Imaging.” American Journal of Roentgenology 211, no. 1: 52–66. [DOI] [PubMed] [Google Scholar]
  3. Al-Mrabeh A, Hollingsworth KG, Shaw JAM, et al. 2020. “2-Year Remission of Type 2 Diabetes and Pancreas Morphology: A Post-Hoc Analysis of the DiRECT Open-Label, Cluster-Randomised Trial.” Lancet Diabetes and Endocrinology 8, no. 12: 939–948. [DOI] [PubMed] [Google Scholar]
  4. Al-Mrabeh A, Hollingsworth KG, Steven S, and Taylor R. 2016. “Morphology of the Pancreas in Type 2 Diabetes: Effect of Weight Loss With or Without Normalisation of Insulin Secretory Capacity.” Diabetologia 59, no. 8: 1753–1759. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Alzaid A, Aideyan O, and Nawaz S. 1993. “The Size of the Pancreas in Diabetes Mellitus.” Diabetic Medicine 10, no. 8: 759–763. [DOI] [PubMed] [Google Scholar]
  6. Bagur AT, Ridgway G, McGonigle J, Brady SM, and Bulte D. 2020. Pancreas Segmentation-Derived Biomarkers: Volume and Shape Metrics in the UK Biobank Imaging Study, 131–142. Springer. [Google Scholar]
  7. Caglar V, Kumral B, Uygur R, Alkoc OA, Ozen OA, and Demirel H. 2014. “Study of Volume, Weight and Size of Normal Pancreas, Spleen and Kidney in Adults Autopsies.” Forensic Medicine and Anatomy Research 2, no. 3: 63–69. [Google Scholar]
  8. Caglar V, Songur A, Yagmurca M, Acar M, Toktas M, and Gonul Y. 2012. “Age-Related Volumetric Changes in Pancreas: A Stereological Study on Computed Tomography.” Surgical and Radiologic Anatomy 34, no. 10: 935–941. [DOI] [PubMed] [Google Scholar]
  9. Cai Z, Chen S, Chen M, et al. 2025. “Effects of Diabetic Ketoacidosis, Islet Cell Function, and Age of Onset on Pancreas Size, Morphology, and Exocrine Function in Children With Type 1 Diabetes: An Abdominal MRI Study.” Journal of Endocrinological Investigation 48: 1889–1901. [DOI] [PubMed] [Google Scholar]
  10. Chaganti S, Robinson JR, Bermudez C, Lasko T, Mawn LA, and Landman BA. 2017. EMR-Radiological Phenotypes in Diseases of the Optic Nerve and Their Association With Visual Function, 373–381. Deep Learn Med Image Anal Multimodal Learn Clin Decis Support. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Chantarojanasiri T, Hirooka Y, Ratanachu-ek T, Kawashima H, Ohno E, and Goto H. 2015. “Evolution of Pancreas in Aging: Degenerative Variation or Early Changes of Disease?” Journal of Medical Ultrasonics 42, no. 2: 177–183. [DOI] [PubMed] [Google Scholar]
  12. D’Antonoli TA, Berger LK, Indrakanti AK, et al. 2024. TotalSegmentator MRI: Sequence-Independent Segmentation of 59 Anatomical Structures in MR Images. RSNA. [Google Scholar]
  13. Desouza SV, Yoon HD, Singh RG, and Petrov MS. 2018. “Quantitative Determination of Pancreas Size Using Anatomical Landmarks and Its Clinical Relevance: A Systematic Literature Review.” Clinical Anatomy 31, no. 6: 913–926. [DOI] [PubMed] [Google Scholar]
  14. Dogan RO, Dogan H, Bayrak C, and Kayikcioglu T. 2021. “A Two-Phase Approach Using Mask R-CNN and 3D U-Net for High-Accuracy Automatic Segmentation of Pancreas in CT Imaging.” Computer Methods and Programs in Biomedicine 207: 106141. [DOI] [PubMed] [Google Scholar]
  15. Fonseca V, Berger LA, Beckett AG, and Dandona P. 1985. “Size of Pancreas in Diabetes Mellitus: A Study Based on Ultrasound.” BMJ 291, no. 6504: 1240–1241. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Fortson BL, Abu-El-Haija M, Mahalingam N, Thompson TL, Vitale DS, and Trout AT. 2024. “Pancreas Volumes in Pediatric Patients Following Index Acute Pancreatitis and Acute Recurrent Pancreatitis.” Pancreatology 24, no. 1: 1–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Garcia TS, Rech TH, and Leitão CB. 2017. “Pancreatic Size and Fat Content in Diabetes: A Systematic Review and Meta-Analysis of Imaging Studies.” PLoS One 12, no. 7: e0180911. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Gilbeau JP, Poncelet V, Libon E, Derue G, and Heller FR. 1992. “The Density, Contour, and Thickness of the Pancreas in Diabetics: CT Findings in 57 Patients.” American Journal of Roentgenology 159, no. 3: 527–531. [DOI] [PubMed] [Google Scholar]
  19. Glaser J, and Stienecker K. 2000. “Pancreas and Aging: A Study Using Ultrasonography.” Gerontology 46, no. 2: 93–96. [DOI] [PubMed] [Google Scholar]
  20. Goda K, Sasaki E, Nagata K, Fukai M, Ohsawa N, and Hahafusa T. 2001. “Pancreatic Volume in Type 1 Und Type 2 Diabetes Mellitus.” Acta Diabetologica 38, no. 3: 145–149. [DOI] [PubMed] [Google Scholar]
  21. Gupta R, Saini S, Sharma S, Jacob TG, and Roy TS. 2017. “Morphology of the Human Pancreas During Development and Aging.” In Topics in Biomedical Gerontology, 67–89. Springer; Singapore. [Google Scholar]
  22. Hastier P, Buckley MJM, Dumas R, et al. 1998. “A Study of the Effect of Age on Pancreatic Duct Morphology.” Gastrointestinal Endoscopy 48, no. 1: 53–57. [DOI] [PubMed] [Google Scholar]
  23. Kawaji Y, Yoshikawa T, Nakagawa K, et al. 2021. “Computed Tomography Findings for Predicting the Future Occurrence of Pancreatic Cancer: Value of Pancreatic Volumetry.” International Journal of Clinical Oncology 26, no. 7: 1304–1313. [DOI] [PubMed] [Google Scholar]
  24. Kerley CI, Chaganti S, Nguyen TQ, et al. 2022. “pyPheWAS: A Phenome-Disease Association Tool for Electronic Medical Record Analysis.” Neuroinformatics 20, no. 2: 483–505. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Kim ME, Gao C, Ramadass K, et al. 2024. Scalable Quality Control on Processing of Large Diffusion-Weighted and Structural Magnetic Resonance Imaging Datasets. PLoS One. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Kipp JP, Olesen SS, Mark EB, Frederiksen LC, Drewes AM, and Frøkjær JB. 2019. “Normal Pancreatic Volume in Adults Is Influenced by Visceral Fat, Vertebral Body Width and Age.” Abdominal Radiology 44, no. 3: 958–966. [DOI] [PubMed] [Google Scholar]
  27. Klöppel G, Löhr M, Habich K, Oberholzer M, and Heitz PU. 1985. “Islet Pathology and the Pathogenesis of Type 1 and Type 2 Diabetes Mellitus Revisited.” Pathology and Immunopathology Research 4, no. 2: 110–125. [DOI] [PubMed] [Google Scholar]
  28. Kreel L, and Sandin B. 1973. “Changes in Pancreatic Morphology Associated With Aging.” Gut 14, no. 12: 962–970. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Le Goallec A, Diai S, Collin S, Prost J-B, Vincent T, and Patel CJ. 2022. “Using Deep Learning to Predict Abdominal Age From Liver and Pancreas Magnetic Resonance Images.” Nature Communications 13, no. 1: 1979. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Li X, Morgan PS, Ashburner J, Smith J, and Rorden C. 2016. “The First Step for Neuroimaging Data Analysis: DICOM to NIfTI Conversion.” Journal of Neuroscience Methods 264: 47–56. [DOI] [PubMed] [Google Scholar]
  31. Löhr J-M, Panic N, Vujasinovic M, and Verbeke CS. 2018. “The Ageing Pancreas: A Systematic Review of the Evidence and Analysis of the Consequences.” Journal of Internal Medicine 283, no. 5: 446–460. [DOI] [PubMed] [Google Scholar]
  32. Macauley M, Percival K, Thelwall PE, Hollingsworth KG, and Taylor R. 2015. “Altered Volume, Morphology and Composition of the Pancreas in Type 2 Diabetes.” PLoS One 10, no. 5: e0126825. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. McCleary BM, Trout AT, Fei L, et al. 2020. “Healthy Pancreatic Parenchymal Volume and Its Relationship to Exocrine Function.” Pediatric Radiology 50, no. 5: 684–688. [DOI] [PubMed] [Google Scholar]
  34. Migdalis IN, Voudouris G, Kalogeropoulou K, Iliopoulou V, Koutoulidis K, and Samartzis M. 1991. “Size of the Pancreas in NonInsulin-Dependent Diabetic Patients.” Journal of Medicine 22, no. 3: 179–186. [PubMed] [Google Scholar]
  35. Möller K, Jenssen C, Braden B, et al. 2023. “Pancreatic Changes With Lifestyle and Age: What Is Normal and What Is Concerning?” Endoscopic Ultrasound 12, no. 2: 213–227. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Pollé OG, Delfosse A, Michoux N, et al. 2023. “Pancreas Imaging of Children With Type 1 Diabetes Reveals New Patterns and Correlations With Pancreatic Functions.” Pediatric Diabetes 2023: 1–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Remedios LW, Liu H, Remedios SW, et al. 2025. “Influence of Early Through Late Fusion on Pancreas Segmentation From Imperfectly Registered Multimodal Magnetic Resonance Imaging.” Journal of Medical Imaging 12, no. 2: 024008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Saini S, Jacob TG, Bhardwaj DN, and Roy TS. 2015. “Age-Related Changes in the Ductular System and Stellate Cells of Human Pancreas.” Journal of the Anatomical Society of India 64, no. 2: 99–106. [Google Scholar]
  39. Saisho Y, Butler AE, Meier JJ, et al. 2007. “Pancreas Volumes in Humans From Birth to Age One Hundred Taking Into Account Sex, Obesity, and Presence of Type-2 Diabetes.” Clinical Anatomy 20, no. 8: 933–942. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Sato T, Ito K, Tamada T, et al. 2012. “Age-Related Changes in Normal Adult Pancreas: MR Imaging Evaluation.” European Journal of Radiology 81, no. 9: 2093–2098. [DOI] [PubMed] [Google Scholar]
  41. Sequeira IR, Yip WC, Lu LWW, et al. 2022. “Pancreas Fat, an Early Marker of Metabolic Risk? A Magnetic Resonance Study of Chinese and Caucasian Women: TOFI_Asia Study.” Frontiers in Physiology 13: 819606. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Shuey MM, Stead WW, Aka I, et al. 2023. “Next-Generation Phenotyping: Introducing phecodeX for Enhanced Discovery Research in Medical Phenomics.” Bioinformatics 39, no. 11: btad655. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Silva MER, Vezozzo DP, Ursich MJM, Rocha DM, Cerri GG, and Wajchenberg BL. 1993. “Ultrasonographic Abnormalities of the Pancreas in IDDM and NIDDM Patients.” Diabetes Care 16, no. 9: 1296–1297. [DOI] [PubMed] [Google Scholar]
  44. Stasinopoulos DM, and Rigby RA. 2007. “Journal of Statistical Software Generalized Additive Models for Location Scale and Shape (GAMLSS) in R.” Journal of Statistical Software 23, no. 7: 1–46. [Google Scholar]
  45. Statsenko Y, Habuza T, Smetanina D, et al. 2023. “Unraveling Lifelong Brain Morphometric Dynamics: A Protocol for Systematic Review and Meta-Analysis in Healthy Neurodevelopment and Ageing.” Biomedicine 11, no. 7: 1999. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Syed A-B, Mahal RS, Schumm LP, and Dachman AH. 2012. “Pancreas Size and Volume on Computed Tomography in Normal Adults.” Pancreas 41, no. 4: 589–595. [DOI] [PubMed] [Google Scholar]
  47. van Griethuysen JJM, Fedorov A, Parmar C, et al. 2017. “Computational Radiomics System to Decode the Radiographic Phenotype.” Cancer Research 77, no. 21: e104–e107. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Vesterhus M, Haldorsen IS, Ræder H, Molven A, and Njølstad PR. 2008. “Reduced Pancreatic Volume in Hepatocyte Nuclear Factor 1A-Maturity-Onset Diabetes of the Young.” Journal of Clinical Endocrinology and Metabolism 93, no. 9: 3505–3509. [DOI] [PubMed] [Google Scholar]
  49. Virostko J, Craddock RC, Williams JM, et al. 2021. “Development of a Standardized MRI Protocol for Pancreas Assessment in Humans.” PLoS One 16, no. 8: e0256029. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Wang M, Gorelick F, and Bhargava A. 2021. “Sex Differences in the Exocrine Pancreas and Associated Diseases.” Cellular and Molecular Gastroenterology and Hepatology 12, no. 2: 427–441. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Wasserthal J, Breit H-C, Meyer MT, et al. 2022. TotalSegmentator: Robust Segmentation of 104 Anatomical Structures in CT Images. RSNA. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Wasserthal J, Breit H-C, Meyer MT, et al. 2023. “TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images.” Radiology: Artifical Intelligence 5, no. 5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Wright JJ, Williams JM, Greeley SAW, Powers AC, Virostko J, and Moore DJ. 2022. “206-OR: Insulin Gene Mutation in Family With Monogenic Diabetes Leads to Smaller Pancreas.” Diabetes 71, no. Supplement_1. [Google Scholar]
  54. Zhou Y, Lee HH, Tang Y, et al. 2025. “Multi-Contrast Computed Tomography Atlas of Healthy Pancreas With Dense Displacement Sampling Registration.” Journal of Medical Imaging 12, no. 02: 024006. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

Research data are not shared.

RESOURCES