Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Unified Framework for Semi-Supervised Image Segmentation and Registration

Ruizhe Li, Grazziela Figueredo, Dorothee Auer, Rob Dineen, Paul Morgan, Xin Chen

TL;DR

The paper tackles the scarcity of annotated medical images for segmentation by coupling an unsupervised image registration model with a segmentation model to produce geometry-preserving pseudo-labels. It introduces a soft pseudo-mask generation mechanism that averages multiple probabilistic masks from both models, enabling iterative joint training that improves both segmentation and registration. The approach yields superior Dice Coefficients and lower Hausdorff distances compared to a Mean Teacher baseline, even when only 1% of annotations are available. The generated soft masks also serve as confidence maps for quality control, highlighting the method's practical potential for robust semi-supervised medical image analysis under data scarcity.

Abstract

Semi-supervised learning, which leverages both annotated and unannotated data, is an efficient approach for medical image segmentation, where obtaining annotations for the whole dataset is time-consuming and costly. Traditional semi-supervised methods primarily focus on extracting features and learning data distributions from unannotated data to enhance model training. In this paper, we introduce a novel approach incorporating an image registration model to generate pseudo-labels for the unannotated data, producing more geometrically correct pseudo-labels to improve the model training. Our method was evaluated on a 2D brain data set, showing excellent performance even using only 1\% of the annotated data. The results show that our approach outperforms conventional semi-supervised segmentation methods (e.g. teacher-student model), particularly in a low percentage of annotation scenario. GitHub: https://github.com/ruizhe-l/UniSegReg.

A Unified Framework for Semi-Supervised Image Segmentation and Registration

TL;DR

The paper tackles the scarcity of annotated medical images for segmentation by coupling an unsupervised image registration model with a segmentation model to produce geometry-preserving pseudo-labels. It introduces a soft pseudo-mask generation mechanism that averages multiple probabilistic masks from both models, enabling iterative joint training that improves both segmentation and registration. The approach yields superior Dice Coefficients and lower Hausdorff distances compared to a Mean Teacher baseline, even when only 1% of annotations are available. The generated soft masks also serve as confidence maps for quality control, highlighting the method's practical potential for robust semi-supervised medical image analysis under data scarcity.

Abstract

Semi-supervised learning, which leverages both annotated and unannotated data, is an efficient approach for medical image segmentation, where obtaining annotations for the whole dataset is time-consuming and costly. Traditional semi-supervised methods primarily focus on extracting features and learning data distributions from unannotated data to enhance model training. In this paper, we introduce a novel approach incorporating an image registration model to generate pseudo-labels for the unannotated data, producing more geometrically correct pseudo-labels to improve the model training. Our method was evaluated on a 2D brain data set, showing excellent performance even using only 1\% of the annotated data. The results show that our approach outperforms conventional semi-supervised segmentation methods (e.g. teacher-student model), particularly in a low percentage of annotation scenario. GitHub: https://github.com/ruizhe-l/UniSegReg.

Paper Structure

This paper contains 12 sections, 3 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overall Framework Architecture
  4. Soft Pseudo-masks Generation
  5. Model Training
  6. Experiments and results
  7. Dataset
  8. Experimental Design and Parameter Settings
  9. Results
  10. Discussion and Conclusions
  11. Compliance with Ethical Standards
  12. Acknowledgments

Figures (2)

  • Figure 1: The proposed joint training framework for one iteration. It consists of soft pseudo-mask generation, segmentation model training, and registration model training. Pseudo-mask generation merges masks from both models for unannotated images to create soft pseudo-masks. These, with annotated images, form a new training set to enhance both models.
  • Figure 2: Visualization results of an example that participated in training as an unannotated image in Joint-1% (row 1-3), MT-1% (row 4) and MT-10% (row 5).