Multi-defect microscopy image restoration under limited data conditions

TL;DR

This paper introduces a unified two-stage deep learning approach using GANs for restoring fluorescence microscopy images with multiple defects under limited training data conditions, outperforming existing methods.

Contribution

A novel two-stage method combining GAN-based data augmentation and conditional GAN restoration tailored for multi-defect microscopy images with scarce data.

Findings

Outperforms CARE, deblurGAN, and CycleGAN with limited data

Effective in restoring multiple defect types

Requires less training data for comparable results

Abstract

Deep learning methods are becoming widely used for restoration of defects associated with fluorescence microscopy imaging. One of the major challenges in application of such methods is the availability of training data. In this work, we propose a unified method for reconstruction of multi-defect fluorescence microscopy images when training data is limited. Our approach consists of two stages: first, we perform data augmentation using Generative Adversarial Network (GAN) with conditional instance normalization (CIN); second, we train a conditional GAN (cGAN) on paired ground-truth and defected images to perform restoration. The experiments on three common types of imaging defects with different amounts of training data show that the proposed method gives comparable results or outperforms CARE, deblurGAN and CycleGAN in restored image quality when available data is limited.