Related Experiment Video
Updated: May 24, 2025
08:40
Quantitation of Protein Expression and Co-localization Using Multiplexed Immuno-histochemical Staining and Multispectral Imaging
Published on: April 8, 2016
12.7K
HisynSeg: Weakly-Supervised Histopathological Image Segmentation via Image-Mixing Synthesis and Consistency
IEEE Transactions on Medical Imaging
|March 3, 2025
Summary
This study introduces HisynSeg, a new method for histopathological image segmentation that uses image synthesis and consistency regularization. HisynSeg improves segmentation accuracy by converting weakly-supervised tasks into fully-supervised ones, overcoming limitations of class activation map methods.
Area of Science:
- Computational Pathology
- Medical Image Analysis
- Computer Vision
Background:
- Tissue semantic segmentation is crucial in computational pathology.
- Class activation map (CAM) methods are used for weakly-supervised segmentation but suffer from activation issues.
- Pixel-level annotations are expensive and time-consuming to acquire.
Purpose of the Study:
- To develop a novel framework for weakly-supervised semantic segmentation of histopathological images.
- To address the under-activation and over-activation problems inherent in CAM-based methods.
- To improve segmentation accuracy without requiring extensive pixel-level annotations.
Main Methods:
- Proposed HisynSeg framework utilizing image-mixing synthesis and consistency regularization.
- Generated synthesized histopathological images with pixel-level masks using Mosaic transformation and Bézier mask generation.
- Implemented an image filtering module for authenticity and self-supervised consistency regularization to prevent overfitting to synthesis artifacts.
Main Results:
- HisynSeg successfully transforms weakly-supervised segmentation into a fully-supervised problem.
- Achieved significant improvements in segmentation accuracy.
- Demonstrated state-of-the-art performance on three independent datasets.
Conclusions:
- The HisynSeg framework effectively enhances semantic segmentation in histopathological images.
- The combination of image synthesis and consistency regularization overcomes limitations of existing weakly-supervised methods.
- The proposed approach offers a more accurate and efficient solution for computational pathology tasks.
