Reference-Free Isotropic 3D EM Reconstruction using Diffusion Models

TL;DR

This paper introduces a diffusion-model-based framework for reference-free isotropic 3D electron microscopy reconstruction, overcoming anisotropic resolution issues without needing prior knowledge or reference data, and demonstrating robustness and self-supervised capabilities.

Contribution

It presents a novel diffusion-model approach that reconstructs 3D EM volumes from highly downsampled data without reference data or prior knowledge, outperforming supervised methods.

Findings

Outperforms supervised learning methods in robustness and quality

Enables self-supervised reconstruction of single anisotropic volumes

Works effectively on public EM datasets

Abstract

Electron microscopy (EM) images exhibit anisotropic axial resolution due to the characteristics inherent to the imaging modality, presenting challenges in analysis and downstream tasks.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.