This repository contains a comprehensive suite of algorithms and tools developed for the processing, reconstruction, and analysis of large-scale electron microscopy (EM) data, with a specific focus on zebrafish connectomics.
The toolkit covers the entire pipeline from raw image quality assessment and denoising to neuron segmentation, organelle detection, and downstream data clustering.
The codebase is organized into the following key components:
Blind2Unblind(EM Denoising v1.0)- The baseline version of our EM image denoising framework. It provides the foundational algorithms for unsupervised noise removal in high-resolution EM datasets.
Blind2Sound(EM Denoising v2.0)- Our advanced, self-supervised denoising algorithm designed specifically for electron microscopy images. This version improves upon previous iterations by effectively suppressing noise without introducing residual artifacts, preserving high-frequency structural details.
Segmentation- Deep learning models for the dense segmentation of zebrafish neurons. This module handles the voxel-wise classification required to isolate individual neuronal structures from background tissue.
Aggregate- Neuron Aggregation Algorithms: A set of post-processing methods used to aggregate over-segmented fragments or probabilistic maps into coherent neuron proposals during the reconstruction process.
Merge_block- Block-wise Stitching: Algorithms designed to solve the large-scale reconstruction challenge. This module handles the seamless stitching and merging of neuron segments that span across different processing blocks (sub-volumes), ensuring global continuity.
Zeb-Mito-Syn- Ultrastructure Recognition: A specialized detection suite for identifying and segmenting intracellular organelles, specifically mitochondria and synapses, within the zebrafish EM volume.
Correlative Light and Electron Microscopy(CLEM)- Registration Framework: Robust algorithms for the cross-modal alignment of Light Microscopy (LM) and Electron Microscopy (EM) datasets, enabling precise overlay of functional signals onto structural data.
Quality Assessment- Automated IQA: Algorithms for the objective assessment of image quality in zebrafish EM acquisitions. This module helps identify artifacts, blur, or contrast issues in raw data before processing.
Zfish-Nc-cluster-all- Clustering & Analysis: A comprehensive analysis package for the final reconstruction results. It includes tools for feature extraction, neuronal clustering, and statistical evaluation of the connectome data.
If you use this code or models in your research, please cite:
- F.-N. Li, J.-Z. Liu, C. Shi, J.-B. Yuan, Y.-N. Lv, J. Liu, L.-N. Zhang, L.-L. Li, L.J. Shen, X. Chen et al., Multiplexed neuromodulatory-type-annotated whole-brain em-reconstruction of larval zebrafish, bioRxiv (2025).