Reconstructing interpretable features in computational super-resolution microscopy via regularized latent search.

Area of Science:

• Microscopy
• Computational Imaging
• Deep Learning

Background:

• Super-resolution (SR) methods artificially increase microscopy image resolution.
• Supervised deep learning requires paired low-resolution/high-resolution images, yielding moderate resolution increases.
• Generative adversarial network (GAN) latent search offers drastic resolution increases without paired images but limited interpretable features.

Purpose of the Study:

• To propose a robust SR method balancing fidelity and realism.
• To enable a two-step analysis: deep learning SR followed by handcrafted quantification.
• To facilitate applications like mobile diagnostics by preserving quantifiable differences.

Main Methods:

• Developed a robust super-resolution (SR) method based on regularized latent search (RLS).
• Integrated deep learning for computational SR with handcrafted algorithms for quantification.
• Utilized a distribution prior for balancing ground truth fidelity and image realism.

Main Results:

• Achieved an actionable balance between fidelity to ground truth and realism of recovered images.
• Enabled splitting image analysis into SR and quantification tasks.
• Demonstrated potential for applications requiring explainable and quantifiable differences.

Conclusions:

• The proposed RLS method offers a robust approach to microscopy image super-resolution.
• This method facilitates accurate quantification of biological features, crucial for diagnostics.
• The two-step analysis process is suitable for resource-limited environments like mobile devices.