Fluorescence microscopy suffers from resolution anisotropy, where axial resolution is significantly lower than lateral resolution. This paper introduces Self-Net, a deep self-learning method that leverages the natural anisotropy to improve axial resolution using lateral images from the same dataset. By combining unsupervised learning for realistic anisotropic degradation and supervised learning for high-fidelity isotropic recovery, Self-Net suppresses hallucination and enhances image quality. Experiments demonstrate its effectiveness across various microscopy platforms, enabling isotropic whole-brain imaging at 0.2 x 0.2 x 0.2 µm³ resolution, significantly improving single-neuron morphology visualization and reconstruction.
