Lifespan Pancreas Morphology for Control vs Type 2 Diabetes using AI on Largescale Clinical Imaging

Lucas W. Remedios, Chloe Cho, Trent M. Schwartz, Dingjie Su, Gaurav Rudravaram, Chenyu Gao, Aravind R. Krishnan, Adam M. Saunders, Michael E. Kim, Shunxing Bao, Thomas A. Lasko, Alvin C. Powers, Bennett A. Landman, John VirostkoarXiv 2025 · DOI · arXiv

Abstract

This study establishes a comprehensive normative reference for pancreatic morphology across the human lifespan (ages 0–90) using automated AI segmentation on a large clinical dataset. By analyzing over 2,500 CT and MRI scans, researchers demonstrate that Type 2 Diabetes (T2D) significantly alters the organ's structure, leading to premature atrophy and distinct morphological changes that deviate from healthy aging trajectories.

Introduction

The human pancreas is a dynamic organ that undergoes significant structural changes, including atrophy and fat infiltration, as part of the natural aging process. Distinguishing between normal age-related decline and disease-induced pathology, such as that seen in Type 2 Diabetes (T2D), has historically been challenging due to the labor-intensive nature of manual organ segmentation in medical imaging. This study addresses the need for a robust, large-scale normative baseline of pancreatic morphology to accurately detect how diseases cause the organ to deviate from its expected aging trajectory.

Methods

An automated pipeline was developed to process a deidentified dataset of 2,533 patients from the Vanderbilt University Medical Center. The researchers utilized TotalSegmentator, a pre-trained deep learning tool, to automatically delineate the pancreas in 3D across both CT and MRI modalities.

From these segmentations, 13 distinct morphological features—including volume, surface area, elongation, and flatness—were extracted using the PyRadiomics library. To account for the non-linear nature of biological aging, the team employed Generalized Additive Models for Location, Scale, and Shape (GAMLSS). This allowed for modeling the average trend and the statistical spread of the data while adjusting for covariates like age, sex, and weight. To isolate the effect of diabetes, a matched cohort of 1,350 patients (675 T2D vs. 675 controls) was analyzed.

Key Results and Insights

The study successfully reproduced known population-level volume trends at a much larger scale than previous manual studies. Pancreatic volume generally increases through early adulthood, peaks in middle age, and then begins a steady decline.

Impact of Type 2 Diabetes

The results strongly reinforce the "smaller pancreas" hypothesis for T2D. After adjusting for age, sex, and weight, 10 out of the 13 morphological features were significantly different (p < 0.05) between T2D patients and non-diabetic controls. T2D essentially shifts the entire volume curve downward, resulting in smaller organs across the adult lifespan.

Comparison of pancreatic morphology across age and disease state. The pancreas typically peaks in volume during middle age before declining; T2D accelerates this reduction in size and alters shape.

Modality Sensitivity

A critical technical finding was the presence of modality bias: MRI tended to yield smaller pancreas measurements than CT when using the same AI tool. This suggests that AI models for organ segmentation require modality-specific baselines or robust harmonization before they can be used interchangeably in clinical research.

Discussion and Future Work

While volume is the most common metric, this study demonstrates that "shape" features like elongation and sphericity provide a deeper characterization of how T2D affects the organ's physical structure.

Limitations of the current work include the 3mm isotropic resolution of the scans, which may be too coarse to detect fine-grained structural changes or subtle "serrated edges." Additionally, the use of contrast-enhanced CT makes accurate fat quantification difficult. Future research utilizing non-contrast imaging or specialized MRI sequences could provide better insights into the interplay between pancreatic fat infiltration and morphological shifts.

Conclusion

By leveraging AI to analyze large-scale clinical data, this research provides a vital reference for pancreatic aging and demonstrates the profound structural impact of Type 2 Diabetes, offering a new framework for identifying pathological deviations in organ morphology.

Summary updated

BibTeX
@misc{remedios2025lifespan,
  author = {Lucas W. Remedios and Chloe Cho and Trent M. Schwartz and Dingjie Su and Gaurav Rudravaram and Chenyu Gao and Aravind R. Krishnan and Adam M. Saunders and Michael E. Kim and Shunxing Bao and Thomas A. Lasko and Alvin C. Powers and Bennett A. Landman and John Virostko},
  title = {{Lifespan Pancreas Morphology for Control vs Type 2 Diabetes using AI on Largescale Clinical Imaging}},
  year = {2025},
  eprint = {2508.14878},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2508.14878},
}
Figure