Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large-scale EM Benchmark for Multi-Organelle Instance Segmentation in the Wild

Yanrui Lu, Danyang Chen, Haowen Xiao, Jiarui Zhu, Fukang Ge, Binqian Zou, Jiali Guan, Jiayin Liang, Yuting Wang, Ziqian Guan, Xiangcheng Bao, Jinhao Bi, Lin Gu, Jun He, Yingying Zhu

TL;DR

This work tackles the insufficiency of small, curated EM benchmarks for multi-organelle instance segmentation by introducing a large-scale, in-the-wild benchmark with over 100k 2D EM images across diverse cell types and five organelle classes. A hybrid Data Refinement pipeline leveraging 3D Connected Component Labeling and expert refinement converts semantic annotations into high-quality instance labels via 3D LPA, enabling robust dataset construction. The authors benchmark three representative segmentation architectures (U-Net, SAM variants, Mask2Former) and reveal a critical mismatch: patch-based methods excel at local organelles but falter on global, network-like structures such as the Endoplasmic Reticulum and Golgi, even after scale alignment. These findings underscore the need for new approaches that integrate long-range context to bridge the gap between real-world heterogeneity and current local-context models. The dataset and labeling tool will be publicly released, providing a resource to drive development in EM instance segmentation.

Abstract

Accurate instance-level segmentation of organelles in electron microscopy (EM) is critical for quantitative analysis of subcellular morphology and inter-organelle interactions. However, current benchmarks, based on small, curated datasets, fail to capture the inherent heterogeneity and large spatial context of in-the-wild EM data, imposing fundamental limitations on current patch-based methods. To address these limitations, we developed a large-scale, multi-source benchmark for multi-organelle instance segmentation, comprising over 100,000 2D EM images across variety cell types and five organelle classes that capture real-world variability. Dataset annotations were generated by our designed connectivity-aware Label Propagation Algorithm (3D LPA) with expert refinement. We further benchmarked several state-of-the-art models, including U-Net, SAM variants, and Mask2Former. Our results show several limitations: current models struggle to generalize across heterogeneous EM data and perform poorly on organelles with global, distributed morphologies (e.g., Endoplasmic Reticulum). These findings underscore the fundamental mismatch between local-context models and the challenge of modeling long-range structural continuity in the presence of real-world variability. The benchmark dataset and labeling tool will be publicly released soon.

Large-scale EM Benchmark for Multi-Organelle Instance Segmentation in the Wild

TL;DR

This work tackles the insufficiency of small, curated EM benchmarks for multi-organelle instance segmentation by introducing a large-scale, in-the-wild benchmark with over 100k 2D EM images across diverse cell types and five organelle classes. A hybrid Data Refinement pipeline leveraging 3D Connected Component Labeling and expert refinement converts semantic annotations into high-quality instance labels via 3D LPA, enabling robust dataset construction. The authors benchmark three representative segmentation architectures (U-Net, SAM variants, Mask2Former) and reveal a critical mismatch: patch-based methods excel at local organelles but falter on global, network-like structures such as the Endoplasmic Reticulum and Golgi, even after scale alignment. These findings underscore the need for new approaches that integrate long-range context to bridge the gap between real-world heterogeneity and current local-context models. The dataset and labeling tool will be publicly released, providing a resource to drive development in EM instance segmentation.

Abstract

Accurate instance-level segmentation of organelles in electron microscopy (EM) is critical for quantitative analysis of subcellular morphology and inter-organelle interactions. However, current benchmarks, based on small, curated datasets, fail to capture the inherent heterogeneity and large spatial context of in-the-wild EM data, imposing fundamental limitations on current patch-based methods. To address these limitations, we developed a large-scale, multi-source benchmark for multi-organelle instance segmentation, comprising over 100,000 2D EM images across variety cell types and five organelle classes that capture real-world variability. Dataset annotations were generated by our designed connectivity-aware Label Propagation Algorithm (3D LPA) with expert refinement. We further benchmarked several state-of-the-art models, including U-Net, SAM variants, and Mask2Former. Our results show several limitations: current models struggle to generalize across heterogeneous EM data and perform poorly on organelles with global, distributed morphologies (e.g., Endoplasmic Reticulum). These findings underscore the fundamental mismatch between local-context models and the challenge of modeling long-range structural continuity in the presence of real-world variability. The benchmark dataset and labeling tool will be publicly released soon.
Paper Structure (12 sections, 2 equations, 5 figures, 2 tables)

This paper contains 12 sections, 2 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Benchmark
  3. Data Acquisition
  4. Data Refine and Annotation
  5. Data Composition and Analysis
  6. Method
  7. Segmentation model
  8. Label converting tool
  9. Experiments
  10. Implementation Details
  11. Main Results
  12. Conclusion

Figures (5)

  • Figure 1: Overview of the mixed label processing pipeline.
  • Figure 2: Composition of the proposed benchmark across data.
  • Figure 3: Distribution of mitochondria instance sizes in the converted BetaSeg subset.
  • Figure 4: Representative 3D meshes of mitochondria and Golgi instances with small, medium, and large volumes.
  • Figure 5: Qualitative visualization of multi-organelle instance labels, where each color denotes a distinct instance.