EMDiffuse: a diffusion-based deep learning method augmenting ultrastructural imaging and volume electron microscopy

Abstract

Electron microscopy (EM) revolutionized the way to visualize cellular ultrastructure. Volume EM (vEM) has further broadened its three-dimensional nanoscale imaging capacity. However, intrinsic trade-offs between imaging speed and quality of EM restrict the attainable imaging area and volume. Isotropic imaging with vEM for large biological volumes remains unachievable. Here we developed EMDiffuse, a suite of algorithms designed to enhance EM and vEM capabilities, leveraging the cutting-edge image generation diffusion model. EMDiffuse demonstrates outstanding denoising and super-resolution performance, generates realistic predictions without unwarranted smoothness, improves predictions resolution by [~]30%, and exhibits excellent transferability by taking only one pair of images to fine-tune. EMDiffuse also pioneers the isotropic vEM reconstruction task, generating isotropic volume similar to that obtained using advanced FIB-SEM even in the absence of isotropic training data. We demonstrated the robustness of EMDiffuse by generating isotropic volumes from six public datasets obtained from different vEM techniques and instruments. The generated isotropic volume enables accurate organelle reconstruction, making 3D nanoscale ultrastructure analysis faster and more accessible and extending such capability to larger volumes. More importantly, EMDiffuse features self-assessment functionalities and guarantees reliable predictions for all tasks. We envision EMDiffuse to pave the way for more in-depth investigations into the intricate subcellular nanoscale structures within large areas and volumes of biological systems.