Related Experiment Video
Updated: May 13, 2026
08:40
Quantitation of Protein Expression and Co-localization Using Multiplexed Immuno-histochemical Staining and Multispectral Imaging
Published on: April 8, 2016
12.4K
SAM2HIPT: a hybrid deep learning framework integrating SAM2 and HIPT with joint loss optimization for
Yu Yao1, Simin Li2, Yangsheng Hu1
1Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650504, People's Republic of China.
Biomedical Physics & Engineering Express
|April 23, 2026
Summary
This study introduces SAM2HIPT, a novel two-stage framework for accurate nucleus segmentation in immunohistochemistry images. It significantly improves cancer diagnosis by enhancing segmentation accuracy and boundary delineation.
Area of Science:
- Digital Pathology
- Computational Biology
- Medical Image Analysis
Background:
- Nucleus segmentation in immunohistochemistry (IHC) images is crucial for cancer diagnosis and treatment assessment.
- Existing methods struggle with staining heterogeneity, cell density, and background interference, limiting accuracy and boundary delineation.
Purpose of the Study:
- To develop an advanced nucleus segmentation framework for improved accuracy and boundary delineation in IHC images.
- To address the limitations of current methods in handling complex pathological image characteristics.
Main Methods:
- A two-stage framework, SAM2HIPT, combining Segment Anything Model 2 (SAM2) for initial segmentation and Hierarchical Image Pyramid Transformer (HIPT) for refinement.
- SAM2's encoder is frozen while its decoder is fine-tuned, extracting local texture, morphology, and boundary information.
- HIPT refines results using multi-scale, multi-level feature fusion for enhanced nuclear structure and boundary consistency.
- A joint loss function ensures collaborative optimization across both stages.
Main Results:
- Achieved Dice coefficients of 0.92 on BCData and 0.91 on DeepLIIF datasets.
- Obtained HD95 boundary error values of 1.05 pixels and 1.10 pixels on the respective datasets.
- Demonstrated superior segmentation performance and robustness compared to state-of-the-art methods.
Conclusions:
- The proposed SAM2HIPT framework significantly advances nucleus segmentation in IHC images.
- The method offers improved accuracy, boundary delineation, and robustness for digital pathology applications.
- SAM2HIPT shows strong potential for enhancing cancer diagnosis and treatment assessment through precise image analysis.
