Related Experiment Video
Updated: Aug 27, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Unsupervised Deep Learning Methods for Biological Image Reconstruction and Enhancement: An overview from a signal
Mehmet Akçakaya1, Burhaneddin Yaman1, Hyungjin Chung2
1Department of Electrical and Computer Engineering, and Center for Magnetic Resonance Research, University of Minnesota, USA.
IEEE Signal Processing Magazine
|October 3, 2022
Summary
Deep learning enhances biological imaging. Unsupervised methods like self-supervised learning and generative models are crucial for image reconstruction and enhancement when reference data is unavailable.
Area of Science:
- Biomedical imaging
- Computational biology
- Machine learning
Background:
- Deep learning excels in biological image reconstruction and enhancement.
- Supervised learning requires matched reference data, which is often difficult to obtain.
- Unsupervised learning approaches are gaining traction due to the lack of paired data.
Purpose of the Study:
- To provide a comprehensive overview of unsupervised deep learning methods for biological imaging.
- To connect these methods to classical inverse problem frameworks.
- To discuss their diverse applications across various imaging modalities.
Main Methods:
- Overview of self-supervised learning and generative models.
- Application of these methods to biological image reconstruction and enhancement.
- Framing these techniques within classical inverse problem theory.
Main Results:
- Demonstrated success of unsupervised deep learning in biological imaging applications.
- Highlighting the versatility of these methods across different microscopy and neuroimaging techniques.
- Establishing a coherent perspective linking deep learning to inverse problems.
Conclusions:
- Unsupervised deep learning, particularly self-supervised and generative models, offers powerful solutions for biological image reconstruction and enhancement.
- These methods overcome the limitations of supervised approaches by not requiring paired reference data.
- The discussed techniques are broadly applicable to electron microscopy, fluorescence microscopy, deconvolution microscopy, optical diffraction tomography, and functional neuroimaging.
