Cryo-ET Segmentation

Unsupervised multi-scale segmentation of cellular cryo-electron tomograms using stable diffusion

Unsupervised Cryo-ET Segmentation with Stable Diffusion

Working with the Xu Lab at Carnegie Mellon University, I am developing cutting-edge methods for unsupervised segmentation and particle picking in cryo-electron tomography (Cryo-ET) data.

Project Overview

Cryo-electron tomography provides 3D visualizations of cellular structures at near-atomic resolution, but analyzing these complex datasets remains challenging. This project focuses on:

  • Unsupervised Segmentation: Developing foundation model-based approaches for multi-scale segmentation without requiring extensive labeled data
  • Particle Picking: Implementing novel methods for macromolecule localization from extremely sparse labels
  • Tomogram Preprocessing: Advanced denoising and preprocessing pipelines for high-resolution structure reconstruction

Key Achievements

🔬 Publication: Submitted to CVPR 2025 - “Unsupervised Multi-scale Segmentation of Cellular Cryo-electron Tomograms with Stable Diffusion Foundation Model”

📄 Accepted: Briefings in Bioinformatics - “Localization of Macromolecules in Crowded Cellular Cryo-electron Tomograms from Extremely Sparse Labels”

Technical Approach

Our approach leverages:

  • Stable diffusion foundation models adapted for 3D biological imaging
  • Self-supervised learning techniques for feature extraction
  • Multi-scale analysis frameworks for hierarchical structure detection
  • Deep learning architectures optimized for volumetric data

Collaboration

Advisor: Dr. Min Xu, Carnegie Mellon University
Duration: June 2024 - Present

This work represents a significant step forward in automated biological image analysis, with potential applications in structural biology, drug discovery, and cellular mechanism understanding.

References