Related Experiment Video
Updated: Nov 15, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
770
Artefact removal in ground truth deficient fluctuations-based nanoscopy images using deep learning.
Suyog Jadhav1, Sebastian Acuña2, Ida S Opstad2
1Indian Institute of Technology (Indian School of Mines), Dhanbad 826004, India.
Biomedical Optics Express
|March 4, 2021
Summary
Deep learning can now denoise challenging nanoscopy images using a novel simulation-supervised training approach. This method overcomes limitations in acquiring real-world data for super-resolution microscopy, enabling better biological sample analysis.
Area of Science:
- Microscopy and Imaging Science
- Computational Biology
- Artificial Intelligence in Life Sciences
Background:
- Deep learning excels at image denoising with supervised data, but this is unavailable for nanoscopy.
- Nanoscopy images, generated via statistical analysis, present unique challenges due to physical constraints and complex data characteristics.
- Existing noise models are insufficient for nanoscopy, hindering direct application of standard deep learning techniques.
Purpose of the Study:
- To develop a robust simulation-supervised deep learning approach for denoising nanoscopy images.
- To address the lack of supervised training datasets in super-resolution optical microscopy.
- To establish a foundation for applying deep learning to nanoscopy in life sciences.
Main Methods:
- Proposed a simulation-supervised training strategy using deep learning auto-encoder architectures.
- Tested the approach on nanoscopy images of sub-cellular structures.
- Investigated generalizability across different nanoscopy methods and biological structures not used in training.
- Evaluated various loss functions and learning models, and analyzed performance metric limitations.
Main Results:
- Demonstrated proof of concept for denoising nanoscopy images using the proposed method.
- Showcased the generalizability of the approach across different nanoscopy techniques and biological samples.
- Provided insights into optimal loss functions and learning models for this specific application.
- Highlighted limitations of current performance metrics for evaluating nanoscopy image quality.
Conclusions:
- The simulation-supervised deep learning approach offers a viable solution for denoising challenging nanoscopy images.
- This work overcomes a significant hurdle in applying AI to super-resolution microscopy.
- The findings pave the way for advanced deep learning applications in nanoscopy for biological research.
