ML-SIM: universal reconstruction of structured illumination microscopy images using transfer learning | JoVE Visualize

ML-SIM: universal reconstruction of structured illumination microscopy images using transfer learning

Charles N Christensen^1,2, Edward N Ward¹, Meng Lu¹

¹University of Cambridge, Department of Chemical Engineering and Biotechnology, Laser Analytics Group, Philippa Fawcett Dr, Cambridge, UK.

Biomedical Optics Express|June 14, 2021

View abstract on PubMed

Summary

A new deep learning method, ML-SIM, reconstructs super-resolution images from structured illumination microscopy (SIM) data faster and more robustly than traditional techniques. This parameter-free approach generalizes across different systems and sample types, improving live-cell imaging.

Area of Science:

Optical microscopy
Super-resolution imaging
Computational imaging

Background:

Structured illumination microscopy (SIM) offers enhanced resolution for live-cell imaging.
Traditional SIM image reconstruction is slow, prone to artifacts, and requires parameter tuning.
Existing methods lack generalizability across different imaging systems and sample types.

Purpose of the Study:

Introduce ML-SIM, a versatile, parameter-free reconstruction method for SIM.
Improve the speed, robustness, and generalizability of SIM image reconstruction.
Enable advanced super-resolution microscopy for challenging imaging conditions.

Main Methods:

Developed ML-SIM, an end-to-end deep residual neural network using transfer learning.
Trained the model on simulated images reflecting challenging real-world conditions.
Applied ML-SIM to diverse sample types and distinct SIM microscopes.

Main Results:

ML-SIM achieves high-quality reconstructions with improved generality and robustness to noise.
Demonstrated successful application on multiple sample types and distinct SIM microscopes.
Reconstruction of a SIM stack is completed in under 200 ms on a GPU.

Conclusions:

ML-SIM offers a robust and versatile alternative to traditional SIM reconstruction methods.
The parameter-free nature and speed enable wider adoption in challenging live-cell imaging.
ML-SIM advances super-resolution microscopy by overcoming limitations of existing reconstruction techniques.