Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Deep Learning Based Structured Illumination Microscopy Reconstruction with Light Field Awareness

Long-Kun Shan, Ze-Hao Wang, Tong-Tian Weng, Xiang-Dong Chen, Fang-Wen Sun

TL;DR

The paper tackles artefacts and robustness issues in deep-learning-based structured illumination microscopy (DL-SIM) that arise when illumination or sample conditions shift from training data. It introduces AL-SIM, a two-stage framework that first estimates the actual light field from raw SIM data and then trains a reconstruction model on bias-corrected synthetic data generated from these predictions, thereby aligning training and test distributions. The approach yields lower reconstruction error and fewer artefacts in both simulated distortions and live-cell imaging of BSC1 cells, with quantitative gains in decorrelation-based and Fourier-domain resolutions. By enhancing consistency between training and real data, AL-SIM broadens SIM’s applicability to complex biological systems and dynamic imaging scenarios.

Abstract

Structured illumination microscopy (SIM) is a pivotal technique for dynamic subcellular imaging in live cells. Conventional SIM reconstruction algorithms depend on accurately estimating the illumination pattern and can introduce artefacts when this estimation is imprecise. Although recent deep learning-based SIM reconstruction methods have improved speed, accuracy, and robustness, they often struggle with out-of-distribution data. To address this limitation, we propose an Awareness-of-Light-field SIM (AL-SIM) reconstruction approach that directly estimates the actual light field to correct for errors arising from data distribution shifts. Through comprehensive experiments on both simulated filament structures and live BSC1 cells, our method demonstrates a 7% reduction in the normalized root mean square error (NRMSE) and substantially lowers reconstruction artefacts. By minimizing these artefacts and improving overall accuracy, AL-SIM broadens the applicability of SIM for complex biological systems.

Enhancing Deep Learning Based Structured Illumination Microscopy Reconstruction with Light Field Awareness

TL;DR

The paper tackles artefacts and robustness issues in deep-learning-based structured illumination microscopy (DL-SIM) that arise when illumination or sample conditions shift from training data. It introduces AL-SIM, a two-stage framework that first estimates the actual light field from raw SIM data and then trains a reconstruction model on bias-corrected synthetic data generated from these predictions, thereby aligning training and test distributions. The approach yields lower reconstruction error and fewer artefacts in both simulated distortions and live-cell imaging of BSC1 cells, with quantitative gains in decorrelation-based and Fourier-domain resolutions. By enhancing consistency between training and real data, AL-SIM broadens SIM’s applicability to complex biological systems and dynamic imaging scenarios.

Abstract

Structured illumination microscopy (SIM) is a pivotal technique for dynamic subcellular imaging in live cells. Conventional SIM reconstruction algorithms depend on accurately estimating the illumination pattern and can introduce artefacts when this estimation is imprecise. Although recent deep learning-based SIM reconstruction methods have improved speed, accuracy, and robustness, they often struggle with out-of-distribution data. To address this limitation, we propose an Awareness-of-Light-field SIM (AL-SIM) reconstruction approach that directly estimates the actual light field to correct for errors arising from data distribution shifts. Through comprehensive experiments on both simulated filament structures and live BSC1 cells, our method demonstrates a 7% reduction in the normalized root mean square error (NRMSE) and substantially lowers reconstruction artefacts. By minimizing these artefacts and improving overall accuracy, AL-SIM broadens the applicability of SIM for complex biological systems.

Paper Structure

This paper contains 5 sections, 2 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Method
  3. Result
  4. Discussion
  5. Conclusion

Figures (5)

  • Figure 1: AL-SIM pipeline. (a) The first training stage. (b) the first test stage. (c) the second training stage. (d) the second test stage
  • Figure 2: AL-SIM network architecture. The network is based on U-net.
  • Figure 3: Experimental setup. (a) Optical path of SIM (b) Formation of different illumination light fields and phase modulation
  • Figure 4: Simulation validation of AL-SIM under light field distortions. (a) Distorted illumination patterns (left), simulated filament structures (middle), and raw SIM images (right). (b) Illumination patterns predicted by the AL-SIM first-stage model. (c) Comparative reconstruction results: (i) AL-SIM, (ii) DL-SIM, (iii) Ground Truth, (iv) Widefield microscopy. Scale bars: 1 µ m (main panels), 200 nm (magnified insets)
  • Figure 5: Comparative experimental results. (a) Comparison between AL-SIM and wide-field microscopy. (b) Imaging of region 1 using AL-SIM, DL-SIM, Conventional SIM, and wide-field microscopy. (c) Spatial resolution comparison along the white dashed line, highlighting that AL-SIM outperforms DL-SIM and conventional SIM in terms of artefacts. (d) Logarithmic Fourier spectra of imaging using AL-SIM, DL-SIM, Conventional SIM, and wide-field microscopy. (e) Imaging of region 1 using different techniques and the resolution improvement along the white dashed line. (f) Imaging of region 2 using different techniques and the resolution improvement along the white dashed line. (a: Scale: 10 $\mu$m; b, c: Scale: 2 $\mu$m)