Applications of deep learning for confocal fluorescence microscopy image quality enhancement using unpaired data

The development of deep learning method and open access to ample public datasets have provided an excellent solution for confocal microscopy image translation, which severs as a stepping stone for microscopic imaging and biomedical research. However, in microscopic imaging, most applications of deep learning algorithms are applied in a supervised manner, which hinders the development of deep learning in microscopy owing to the need for large paired images and annotations. Here, we use an improved Cycle Generative Adversarial Network-based unsupervised learning algorithm to achieve confocal microscopy image quality enhancement. Two datasets, the self-collected human prostate cancer cells deblurring dataset and public planaria denoising dataset, are used to train and test in the mentioned model. The results show the model has excellent generalization ability, which can be adapted to different image translation tasks and different types of cell imaging for confocal microscopy.