Architectural order identification across label-free living cell imaging with a swin transformer-conditional GAN.

Abstract

Quantitative Label-Free Imaging Phase and Polarization (QLIPP) technology enables non-invasive analysis and characterization of samples based on their intrinsic properties, without the need for exogenous labeling or contrast agents. However, QLIPP often involves dealing with complex tissue structures, such as overlapping or interconnected regions, making it challenging to accurately depict such intricate architectures. In order to elucidate the inherent ordered structures across spatial and temporal scales in living systems, we propose an efficient architecture based on the Swin Transformer Conditional Generative Adversarial Network (ST-cGAN). This model synergistically combines polarized light microscopy and the cooperative reconstruction of complementary optical properties. Leveraging complementary contrast information, the ST-cGAN achieves high-precision predictions of specific structures, addressing the difficulty of QLIPP in portraying complex tissue structures accurately. We demonstrate the efficacy of the model by predicting ordered structures within different components of kidney tissue morphology, including F-actin and cell nuclei. To enhance the accessibility and reproducibility of our proposed method, the open-source code used for neural network training is available on GitHub. This work marks a significant advancement in the field of label-free live cell imaging, particularly in the identification of ordered structures, contributing to a deeper understanding of dynamic biological processes.