Parameter-Efficient Adaptation of SAM-Med3D for Pancreatic Segmentation: Research Software Scaffold

Overview

This project explores adapting the SAM-Med3D foundation model to three-dimensional pancreas and pancreatic-tumour segmentation. The public software archive contains source code, configuration, tests, and research documentation.

Research question

How can parameter-efficient adaptation of a pretrained 3D segmentation model support pancreas and pancreatic-tumour segmentation under limited training data?

Methods

The documented workflow includes data preprocessing, fixed dataset splits, and a configurable 3D U-Net scaffold. Comparisons with pretrained SAM-Med3D inference, full fine-tuning, and parameter-efficient adaptation are planned.

Data & evaluation setup

Medical Segmentation Decathlon Task07 Pancreas. Medical images and pretrained model weights are not included in the software archive; obtain them from their original providers under the applicable terms.

Limitations

The DOI identifies research software, not a manuscript or a peer-reviewed publication. Completed comparative experiments and validated segmentation results are not claimed.

Collaborators & acknowledgements

SAM-Med3D and the Medical Segmentation Decathlon retain their original authorship and licensing terms.