Related Experiment Video
Updated: Nov 27, 2025
Analyzing Mitochondrial Morphology Through Simulation Supervised Learning
Published on: March 3, 2023
MitoEM Dataset: Large-scale 3D Mitochondria Instance Segmentation from EM Images
Donglai Wei1, Zudi Lin1, Daniel Franco-Barranco2,3
1Harvard University.
Abstract:
Electron microscopy (EM) allows the identification of intracellular organelles such as mitochondria, providing insights for clinical and scientific studies. However, public mitochondria segmentation datasets only contain hundreds of instances with simple shapes. It is unclear if existing methods achieving human-level accuracy on these small datasets are robust in practice. To this end, we introduce the MitoEM dataset, a 3D mitochondria instance segmentation dataset with two (30μm)3 volumes from human and rat cortices respectively, 3, 600× larger than previous benchmarks. With around 40K instances, we find a great diversity of mitochondria in terms of shape and density. For evaluation, we tailor the implementation of the average precision (AP) metric for 3D data with a 45× speedup. On MitoEM, we find existing instance segmentation methods often fail to correctly segment mitochondria with complex shapes or close contacts with other instances. Thus, our MitoEM dataset poses new challenges to the field. We release our code and data: https://donglaiw.github.io/page/mitoEM/index.html.
Insights
The new MitoEM dataset, 3,600x larger than previous benchmarks, reveals limitations in current mitochondria segmentation methods. This 3D dataset challenges algorithms with diverse, complex mitochondria shapes and densities found in human and rat cortices.
Area of Science:
- Cell Biology
- Neuroscience
- Computer Vision
Background:
- Electron microscopy (EM) is crucial for identifying intracellular organelles like mitochondria, aiding clinical and scientific research.
- Existing public mitochondria segmentation datasets are limited, containing few instances with simple shapes, raising questions about the robustness of current segmentation methods.
- The complexity and diversity of mitochondria in biological tissues are not adequately represented in current benchmark datasets.
Purpose of the Study:
- To introduce the MitoEM dataset, a large-scale 3D instance segmentation dataset for mitochondria.
- To evaluate the performance of existing instance segmentation methods on a more challenging and representative dataset.
- To highlight the limitations of current methods in segmenting diverse and complex mitochondrial structures.
Main Methods:
- Development of the MitoEM dataset, comprising two (30μm)³ volumes from human and rat cortices, featuring approximately 40,000 mitochondria instances.
- Adaptation of the average precision (AP) metric for 3D data, achieving a 45x speedup for evaluation.
- Testing of existing instance segmentation algorithms on the MitoEM dataset.
Main Results:
- The MitoEM dataset demonstrates a significant diversity in mitochondria shape and density compared to previous benchmarks.
- Existing instance segmentation methods exhibit failures in accurately segmenting mitochondria with complex morphologies or those in close proximity to other organelles.
- Performance analysis revealed that current methods struggle with the scale and complexity present in the MitoEM dataset.
Conclusions:
- The MitoEM dataset presents a substantial advancement in scale and complexity for mitochondria instance segmentation research.
- Current segmentation techniques require further development to achieve robust performance on diverse and complex biological structures.
- The dataset and associated code are released to facilitate future research and development in the field.
