Papers › PanSAM: Zero-Shot, Prompt-Free Pancreas Segmentation in CT Imaging
PanSAM: Zero-Shot, Prompt-Free Pancreas Segmentation in CT Imaging
Abolfazl malekahmadi, Mohammad Taha Teimuri Jervakani, Armin Behnamnia, Zahra Dehghanian, Amir Shamloo, Hamid R. Rabiee
Segmentation of the pancreas in CT images is crucial in multiple pancreatic diagnostic tasks, such as the detection, classification, and prognosis of pancreatic cancer. We present a segmentation model to find pancreatic tissue accurately in abdominal CT images. We utilize the Segment-Anything Model (SAM), a prompt-based 2D segmentation transformer model, and adapt it to 3D CT images to build a model that can segment the pancreas automatically without any prompts. To our knowledge, this is the first prompt-free work to segment the pancreas on a CT image based on the generalizable SAM model. We achieve a DICE score of 87.01% and a Jaccard score of 81.42% on the NIH dataset. We also performed zero-shot segmentation on the Abdominal-1K dataset. We achieved a DICE score of 83.20%, which shows the generalizability and applicability of our method to new unseen samples. Our study put together the zero-shot performance of SAM and the 3D nature of CT images to provide an automatic, real-time model that provides consistent segmentation throughout CT slices without the need for expert intervention.
Code
Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.
Code Syntology ran Syntology
Not run by Syntology. Nothing on this page verifies that the listed code works.
Tasks
Results from the paper archive 2025-07-28
| Task | Dataset | Model | Metric | Value | Rank at snapshot | Leaderboard | Report |
|---|---|---|---|---|---|---|---|
| Pancreas Segmentation | Pancreas-CT | PanSAM | Dice | 87.01 | #1 of 1 | Archive leaderboard | report |
Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.
Methods
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections