PAM: a propagation-based model for segmenting any 3D objects across multi-modal medical images - Takeaways - MDSpire

PAM: a propagation-based model for segmenting any 3D objects across multi-modal medical images

  • By

  • Zifan Chen

  • Xinyu Nan

  • Jiazheng Li

  • Jie Zhao

  • Haifeng Li

  • Ziling Lin

  • Haoshen Li

  • Heyun Chen

  • Yiting Liu

  • Lei Tang

  • Li Zhang

  • Bin Dong

  • December 2, 2025

  • 0 min

Share

  • 1

    PAM is a novel framework for 3D segmentation in medical imaging, utilizing minimal 2D prompts to generate accurate volumetric segmentations.

  • 2

    The framework combines a CNN-based UNet for intra-slice features with Transformer attention for effective inter-slice propagation.

  • 3

    PAM outperformed existing models like MedSAM and SegVol, achieving a 19.3% improvement in average Dice Similarity Coefficient across 44 datasets.

  • 4

    The model demonstrated stable performance under varying prompts and settings, with a significant reduction in user interaction time by 63.6%.

  • 5

    PAM's design reduces reliance on extensive manual annotations, making it a more efficient and generalizable tool for automated clinical imaging.

Original Source(s)

PAM: a propagation-based model for segmenting any 3D objects across multi-modal medical images

  • by Zifan Chen, Xinyu Nan, Jiazheng Li, Jie Zhao, Haifeng Li, Ziling Lin, Haoshen Li, Heyun Chen, Yiting Liu, Lei Tang, Li Zhang, Bin Dong

  • December 2, 2025

Related Content

Deep learning model for assessing survival benefits in hepatocellular carcinoma patients undergoing intra-arterial therapies based on proliferative subtype

An end-to-end deep learning pipeline for hematoma expansion prediction in spontaneous intracerebral hemorrhage based on non-contrast computed tomography

A generalizable 3D framework and model for self-supervised learning in medical imaging