Reference-Free Isotropic 3D EM Reconstruction Using Diffusion Models

Electron microscopy (EM) images exhibit anisotropic axial resolution due to the characteristics inherent to the imaging modality, presenting challenges in analysis and downstream tasks. Recently proposed deep-learning-based isotropic reconstruction methods have addressed this issue; however, training the deep neural networks require either isotropic ground truth volumes, prior knowledge of the degradation process, or point spread function (PSF). Moreover, these methods struggle to generate realistic volumes when confronted with high scaling factors (e.g. \(\times \) 8, \(\times \) 10). In this paper, we propose a diffusion-model-based framework that overcomes the limitations of requiring reference data or prior knowledge about the degradation process. Our approach utilizes 2D diffusion models to consistently reconstruct 3D volumes and is well-suited for highly downsampled data. Extensive experiments conducted on two public datasets demonstrate the robustness and superiority of leveraging the generative prior compared to supervised learning methods. Additionally, we demonstrate our method’s feasibility for self-supervised reconstruction, which can restore a single anisotropic volume without any training data. The source code is available on GitHub: https://github.com/hvcl/diffusion-em-recon .