Three-dimensional (3D) optical microscopy is indispensable for visualizing the structural, functional, and dynamic organization of biological systems, yet its performance is constrained by diffraction, optical aberrations, phototoxicity, limited depth of field, and the growing complexity of volumetric data analysis. Recent advances in deep learning (DL) have profoundly reshaped computational 3D microscopy by enabling data-driven enhancement, reconstruction, and interpretation of volumetric images. This review provides a comprehensive overview of DL-enabled methods and representative results in 3D optical microscopy. We first summarize representative 3D microscopy modalities and their inherent limitations. We then review DL approaches for image quality improvement, including denoising, super-resolution, and depth-of-field extension, followed by DL-driven image generation and reconstruction from sparse or ill-posed measurements. Finally, we discuss DL-based automated image analysis for segmentation and quantitative interpretation, and outline current challenges and future directions.
Zhang et al. (Fri,) studied this question.