Correction of out-of-focus microscopic images by deep learning.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Microscopy

Background:

• High-quality microscopy images are crucial for biomedical research, disease diagnosis, and medical discovery.
• Out-of-focus images degrade performance in research and diagnostics, necessitating advanced correction techniques.

Purpose of the Study:

• To develop and validate an advanced computational model for correcting out-of-focus microscopy images.
• To enhance the quality and utility of microscopic data for scientific and diagnostic applications.

Main Methods:

• A Cycle Generative Adversarial Network (CycleGAN) based model was developed.
• A multi-component weighted loss function was integrated into the CycleGAN model.
• The model was trained and tested on self-collected datasets (Leishmania parasite, BPAEC) and a public dataset (human cells).

Main Results:

• The proposed CycleGAN model achieved state-of-the-art performance in correcting out-of-focus microscopy images compared to existing GAN-based methods.
• The model demonstrated excellent generalization capabilities, successfully transferring to different types of microscopic image datasets.
• The developed model effectively addresses the challenge of blurry images in microscopy.

Conclusions:

• The CycleGAN-based model offers a robust solution for deblurring microscopy images, enhancing their diagnostic and research value.
• The model's strong generalization ability makes it applicable across diverse microscopic imaging scenarios.
• Public availability of code and datasets facilitates further research and application in the field.