cerebellum

This repository contains tools to generate segmentation of parallel fibers from electron microscopy images of neuronal tissue. In addition to scripts to run the traditional connectomics pipeline (affnity generation, region agglomeration), it provides tools for:

• Error analysis based on voxel-wise comparison and skeleton comparison
• Segmentation stitching using IoU based object-tracking
• Basic error correction for false splits and false merges

Getting started

Clone the repository.

git clone --recursive https://github.com/srujanm/cerebellum.git

Clone the ibex repository which contains skeletonization tools.

cd cerebellum/
git clone --recursive https://github.com/srujanm/ibex.git

Follow the installation instructions on the ibex page.

Depending on your requirement, install packages from our traditional EM pipeline.

• EM-network for affinity prediction
• waterz for region agglomeration
• EM-segLib for error analysis

Scripts

The scripts folder is organized into various sub-folders based on the stage of operation in the pipeline. Here are quick descriptions of them from start to end of the pipeline.

• network: training and testing of affinity generation networks
• waterz: region agglomeration from affinity and generation of segmentation
• gt_prep: preparation of ground truth segmentation blocks and skeletons
• seg_prep: preparation and error analysis of predicted segmentation blocks
• seg_prep_unetfiber: preparation and error analysis of predicted segmentation blocks from new affinity network trials
• blockchain: linking of segmentation blocks and assembly of full volume
• tracking: basic error correction
• pnj_scripts: scripts used to refine Purkinje cell segmentation

Notebooks

The notebooks provide tutorial-style introductions to various modules in the repository.

• Skeleton generation
• Block level error analysis
• Block stitching error analysis
• Scaling of errors with tissue volume
• False merge correction
• Comparison of affinity networks