Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Weakly-Supervised Cross-Domain Segmentation of Electron Microscopy with Sparse Point Annotation

Dafei Qiu, Shan Xiong, Jiajin Yi, Jialin Peng

TL;DR

This paper addresses cross-domain organelle segmentation in electron microscopy under extremely weak supervision by introducing WDA-Net, a multi-task pyramid framework that jointly learns segmentation, center detection, and counting. A counting-based global prior guides center detection, while a cross-position cut-and-paste augmentation and entropy-based pseudo-labeling alleviate severe label sparsity, enabling strong cross-domain performance with only sparse target annotations (e.g., 15%). Comprehensive experiments across multiple EM datasets show substantial gains over state-of-the-art unsupervised domain adaptation methods, achieving near-supervised accuracy with far reduced annotation effort. The results suggest significant practical impact for scalable, annotation-efficient segmentation in biomedical imaging, though the approach relies on access to source data/model and is tailored to densely distributed objects like mitochondria.

Abstract

Accurate segmentation of organelle instances from electron microscopy (EM) images plays an essential role in many neuroscience researches. However, practical scenarios usually suffer from high annotation costs, label scarcity, and large domain diversity. While unsupervised domain adaptation (UDA) that assumes no annotation effort on the target data is promising to alleviate these challenges, its performance on complicated segmentation tasks is still far from practical usage. To address these issues, we investigate a highly annotation-efficient weak supervision, which assumes only sparse center-points on a small subset of object instances in the target training images. To achieve accurate segmentation with partial point annotations, we introduce instance counting and center detection as auxiliary tasks and design a multitask learning framework to leverage correlations among the counting, detection, and segmentation, which are all tasks with partial or no supervision. Building upon the different domain-invariances of the three tasks, we enforce counting estimation with a novel soft consistency loss as a global prior for center detection, which further guides the per-pixel segmentation. To further compensate for annotation sparsity, we develop a cross-position cut-and-paste for label augmentation and an entropy-based pseudo-label selection. The experimental results highlight that, by simply using extremely weak annotation, e.g., 15\% sparse points, for model training, the proposed model is capable of significantly outperforming UDA methods and produces comparable performance as the supervised counterpart. The high robustness of our model shown in the validations and the low requirement of expert knowledge for sparse point annotation further improve the potential application value of our model.

Weakly-Supervised Cross-Domain Segmentation of Electron Microscopy with Sparse Point Annotation

TL;DR

This paper addresses cross-domain organelle segmentation in electron microscopy under extremely weak supervision by introducing WDA-Net, a multi-task pyramid framework that jointly learns segmentation, center detection, and counting. A counting-based global prior guides center detection, while a cross-position cut-and-paste augmentation and entropy-based pseudo-labeling alleviate severe label sparsity, enabling strong cross-domain performance with only sparse target annotations (e.g., 15%). Comprehensive experiments across multiple EM datasets show substantial gains over state-of-the-art unsupervised domain adaptation methods, achieving near-supervised accuracy with far reduced annotation effort. The results suggest significant practical impact for scalable, annotation-efficient segmentation in biomedical imaging, though the approach relies on access to source data/model and is tailored to densely distributed objects like mitochondria.

Abstract

Accurate segmentation of organelle instances from electron microscopy (EM) images plays an essential role in many neuroscience researches. However, practical scenarios usually suffer from high annotation costs, label scarcity, and large domain diversity. While unsupervised domain adaptation (UDA) that assumes no annotation effort on the target data is promising to alleviate these challenges, its performance on complicated segmentation tasks is still far from practical usage. To address these issues, we investigate a highly annotation-efficient weak supervision, which assumes only sparse center-points on a small subset of object instances in the target training images. To achieve accurate segmentation with partial point annotations, we introduce instance counting and center detection as auxiliary tasks and design a multitask learning framework to leverage correlations among the counting, detection, and segmentation, which are all tasks with partial or no supervision. Building upon the different domain-invariances of the three tasks, we enforce counting estimation with a novel soft consistency loss as a global prior for center detection, which further guides the per-pixel segmentation. To further compensate for annotation sparsity, we develop a cross-position cut-and-paste for label augmentation and an entropy-based pseudo-label selection. The experimental results highlight that, by simply using extremely weak annotation, e.g., 15\% sparse points, for model training, the proposed model is capable of significantly outperforming UDA methods and produces comparable performance as the supervised counterpart. The high robustness of our model shown in the validations and the low requirement of expert knowledge for sparse point annotation further improve the potential application value of our model.
Paper Structure (15 sections, 13 equations, 9 figures, 4 tables)

This paper contains 15 sections, 13 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Segmentation Head
  5. Detection Head with Partial Point Supervision
  6. Counting Head as a Global Prior
  7. Cross-Position Cut-and-Paste Label Augmentation
  8. Overall Optimization
  9. Experiments
  10. Dataset and Validation Settings
  11. Model Analysis
  12. Comparative Experiments
  13. Visualization of Aligned Feature
  14. Annotation Efficiency
  15. Conclusions and Discussions

Figures (9)

  • Figure 1: Typical EM images scanned from different tissues/species, which result in large domain gaps.
  • Figure 2: Illustration of the sparse point annotation and other types of annotations for mitochondria in EM images. Sparse point annotation not only leads to a significantly reduced annotation workload but also requires much less expert knowledge.
  • Figure 3: Performance of source models without domain adaptation on the target domain. Red: ground truth annotations; Green: predicted center points/segmentation. a) A source image from Drosophila III VNC data DrosophilaIIIVNC; b) a target image from EPFL data lucchi2013learning; c) predicted counts by the source counting model; d) predicted center-points by the source detection model; e) predicted segmentation by the source segmentation model.
  • Figure 4: Illustration of our proposed WDA-Net, which jointly learns three correlated tasks that have different levels of domain invariance. An auxiliary counting task is employed to guide the detection task, which helps recognize mitochondria instances and reduce false positives in both implicit and explicit ways.
  • Figure 5: Source annotation refinement (SAR) by GAC. Yellow: original annotation; Red: refined annotation by GAC. Note that the refinement is only used for source model training but not used for target data.
  • ...and 4 more figures