Related Experiment Video
Updated: Jun 11, 2026
08:47
Quantitative Approaches for Studying Cellular Structures and Organelle Morphology in Caenorhabditis elegans
Published on: July 5, 2019
10.4K
MitoEM 2.0: A Benchmark for Challenging 3D Mitochondria Instance Segmentation from EM Images
Peng Liu1, Boyu Shen1, Liyuan Liu1
1Boston College.
Biorxiv : the Preprint Server for Biology
|February 25, 2026
Summary
MitoEM 2.0 is a new dataset for training 3D mitochondria segmentation in electron microscopy. It features expert labels for complex scenarios, enabling robust algorithm development and benchmarking.
Area of Science:
- Cell Biology
- Biophysics
- Computer Vision
Background:
- Accurate 3D mitochondria segmentation in volume electron microscopy (vEM) is crucial for understanding cellular function.
- Existing datasets often lack the complexity and expert annotation required for robust model training and evaluation.
Purpose of the Study:
- To introduce MitoEM 2.0, a comprehensive dataset for training and evaluating 3D mitochondria instance segmentation algorithms.
- To provide standardized, expert-verified data covering challenging biological scenarios.
Main Methods:
- Assembled multiscale vEM datasets (FIB-SEM, SBF-SEM, ssSEM) from diverse tissues and species.
- Generated expert-verified instance labels focusing on dense packing, fused networks, and ambiguous boundaries.
- Provided native and processed volumes, metadata, standardized splits, and NIfTI format.
Main Results:
- MitoEM 2.0 includes challenging datasets with high-quality, expert annotations.
- The dataset enables reproducible benchmarking of segmentation methods.
- Baseline scripts and size-stratified evaluation are provided.
Conclusions:
- MitoEM 2.0 facilitates robust model development for 3D mitochondria segmentation.
- The resource supports fair comparison across different algorithms.
- It serves as a valuable tool for bioimage analysis, algorithm benchmarking, and education.
