Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging Anatomical Priors for Automated Pancreas Segmentation on Abdominal CT

Anisa V. Prasad, Tejas Sudharshan Mathai, Pritam Mukherjee, Jianfei Liu, Ronald M. Summers

TL;DR

The paper addresses the challenge of accurate pancreas segmentation on CT by incorporating anatomical priors. It compares two 3D nnU-Net models trained with and without priors, using PANORAMA-derived labels and TotalSegmentator-based priors, and evaluates on AMOS22. The priors-driven model achieves higher Dice similarity and lower Hausdorff distance, with zero detection failures, demonstrating the potential of anatomical priors to improve automated biomarker extraction from CT. The approach offers a more reliable pancreas segmentation pipeline that could enhance early diagnosis and biomarker studies for pancreatic diseases, while highlighting limitations related to CT protocol variability and vascular labeling. Overall, anatomical priors materially improve segmentation accuracy and robustness in this setting.

Abstract

An accurate segmentation of the pancreas on CT is crucial to identify pancreatic pathologies and extract imaging-based biomarkers. However, prior research on pancreas segmentation has primarily focused on modifying the segmentation model architecture or utilizing pre- and post-processing techniques. In this article, we investigate the utility of anatomical priors to enhance the segmentation performance of the pancreas. Two 3D full-resolution nnU-Net models were trained, one with 8 refined labels from the public PANORAMA dataset, and another that combined them with labels derived from the public TotalSegmentator (TS) tool. The addition of anatomical priors resulted in a 6\% increase in Dice score ($p < .001$) and a 36.5 mm decrease in Hausdorff distance for pancreas segmentation ($p < .001$). Moreover, the pancreas was always detected when anatomy priors were used, whereas there were 8 instances of failed detections without their use. The use of anatomy priors shows promise for pancreas segmentation and subsequent derivation of imaging biomarkers.

Leveraging Anatomical Priors for Automated Pancreas Segmentation on Abdominal CT

TL;DR

The paper addresses the challenge of accurate pancreas segmentation on CT by incorporating anatomical priors. It compares two 3D nnU-Net models trained with and without priors, using PANORAMA-derived labels and TotalSegmentator-based priors, and evaluates on AMOS22. The priors-driven model achieves higher Dice similarity and lower Hausdorff distance, with zero detection failures, demonstrating the potential of anatomical priors to improve automated biomarker extraction from CT. The approach offers a more reliable pancreas segmentation pipeline that could enhance early diagnosis and biomarker studies for pancreatic diseases, while highlighting limitations related to CT protocol variability and vascular labeling. Overall, anatomical priors materially improve segmentation accuracy and robustness in this setting.

Abstract

An accurate segmentation of the pancreas on CT is crucial to identify pancreatic pathologies and extract imaging-based biomarkers. However, prior research on pancreas segmentation has primarily focused on modifying the segmentation model architecture or utilizing pre- and post-processing techniques. In this article, we investigate the utility of anatomical priors to enhance the segmentation performance of the pancreas. Two 3D full-resolution nnU-Net models were trained, one with 8 refined labels from the public PANORAMA dataset, and another that combined them with labels derived from the public TotalSegmentator (TS) tool. The addition of anatomical priors resulted in a 6\% increase in Dice score () and a 36.5 mm decrease in Hausdorff distance for pancreas segmentation (). Moreover, the pancreas was always detected when anatomy priors were used, whereas there were 8 instances of failed detections without their use. The use of anatomy priors shows promise for pancreas segmentation and subsequent derivation of imaging biomarkers.

Paper Structure

This paper contains 7 sections, 3 figures, 3 tables.

Table of Contents

  1. Purpose
  2. METHODS
  3. Results
  4. Discussion
  5. Acknowledgements
  6. Appendix
  7. Model Details

Figures (3)

  • Figure 1: Fully automated framework for pancreas segmentation on CT. Two 3D nnU-Net models were trained with and without anatomy priors. Model 1 (top) was trained with refined PANORAMA dataset labels, while Model 2 (bottom) was trained with the refined labels and anatomy priors (constraints used to guide segmentation based on anatomical structures) obtained from the public TotalSegmentator (TS) tool. Both models were evaluated on the AMOS22 dataset.
  • Figure 2: Standards for Reporting Diagnostic Accuracy (STARD) Chart describing the inclusion and exclusion criteria of patients for the (a) PANORAMA dataset (training) and (b) external AMOS22 dataset (testing).
  • Figure 3: Segmentation output for two representative scans. (a, b) Axial slices of the two scans. (c, d) Reference pancreas segmentations from the AMOS22 dataset. Segmented pancreas from (e, f) TotalSegmentator, (g, h) Model 1 (REF$\_$8), and (i, j) Model 2 (ALL$\_$45).