3D Mitochondria Shape Library for Optical Microscopy (3DMSL): A multimodal dataset for deep learning based...

Area of Science:

• Cell Biology
• Microscopy
• Deep Learning

Background:

• Supervised deep learning for fluorescence microscopy image analysis requires extensive annotated ground truth datasets.
• Acquiring these annotations is time-consuming and costly, posing a significant bottleneck.

Purpose of the Study:

• To address the need for annotated data in fluorescence microscopy, we present 3DMSL, a comprehensive database of 3D mitochondria shapes.
• This database aims to facilitate the training of deep learning models for various image analysis tasks.

Main Methods:

• 3DMSL was created using high-resolution Electron Microscopy data.
• A physics-based simulator was employed to generate large fluorescence microscope image datasets with 3D ground truths from the electron microscopy data.

Main Results:

• The 3DMSL database contains over 27,000 diverse instances of mitochondria shapes.
• These shapes are available in multiple 3D representations, including meshes, point clouds, and implicit shapes.

Conclusions:

• 3DMSL provides a valuable resource for training deep learning models in fluorescence microscopy.
• Applications include image segmentation, 3D reconstruction, time-lapse video generation, and microscope translation.