Improving correlation based super-resolution microscopy images through image fusion by self-supervised deep learning.

Area of Science:

• Microscopy and Imaging Technologies
• Computational Biology and Bioinformatics
• Cell Biology

Background:

• Super-resolution imaging provides high-detail subcellular structure visualization.
• Traditional intensity microscopy offers good signal-to-noise ratio (SNR).
• Super-resolution optical fluctuation imaging (SOFI) achieves high resolution but can have lower SNR.

Purpose of the Study:

• To develop a deep learning approach for fusing intensity and SOFI microscopy images.
• To combine the high resolution of SOFI with the high SNR of intensity images.
• To create a self-supervised network capable of generalizing across different image pairs.

Main Methods:

• A novel deep learning network was constructed for image fusion.
• The network integrates intensity and SOFI microscopy data.
• Self-supervised learning enabled training on single image pairs.

Main Results:

• The fused images achieved resolution comparable to SOFI and SNR comparable to intensity images.
• Experimental validation on microtubule-stained cell samples demonstrated effectiveness.
• The self-supervised network generalized well to new image pairs without retraining.

Conclusions:

• The deep learning approach successfully fuses intensity and SOFI microscopy images.
• This method offers a flexible and efficient way to enhance biological imaging.
• The technique is applicable to other fluctuation-based imaging modalities.