Isotropic multi-scale neuronal reconstruction from high-ratio expansion microscopy with contrastive unsupervised deep
Gary Han Chang1, Meng-Yun Wu2, Ling-Hui Yen3
1Institute of Medical Device and Imaging, College of Medicine, National Taiwan University, Taipei, Taiwan, ROC; Graduate School of Advanced Technology, National Taiwan University, Taipei, Taiwan, ROC.
Computer Methods and Programs in Biomedicine
|January 7, 2024
Summary
IsoGAN, an unsupervised generative adversarial network, reconstructs complex 3D neuronal structures from expansion microscopy data. This method overcomes limitations in resolution and manual annotation for high-throughput morphology analysis.
Area of Science:
- Neuroscience
- Biotechnology
- Medical Imaging
Background:
- Expansion microscopy (ExM) data analysis is hindered by anisotropic optical resolution.
- Manual annotation of complex neuronal structures presents a significant bottleneck.
Purpose of the Study:
- To develop an innovative imaging reconstruction approach for high-ratio ExM data.
- To overcome limitations of current methods in analyzing complex neuronal structures.
Main Methods:
- Devised the IsoGAN model, a contrastive unsupervised generative adversarial network.
- Leveraged multi-scale and isotropic morphology data for 3D structure generation.
- Utilized simplified structures with idealized morphologies as shape priors.
Main Results:
- IsoGAN generates high-fidelity 3D representations of neurons, proteins, and blood vessels.
- Eliminated the need for rigorous manual annotation and supervision.
- Accurate reconstruction demonstrated by consistency between axial/lateral views and reduced artifacts.
Conclusions:
- IsoGAN streamlines imaging reconstruction for high-throughput morphology analysis.
- Enables detailed 3D reconstruction using fewer axial view images.
- Offers a transformative solution for analyzing diverse biological structures.
