LUMIN: Light-sheet Microscopy Analysis Unified with Distributed and Domain-Randomized Generative Models.

This repository provides a framework to perform large-scale, distributed image segmentation of light-sheet microscopy (LSM) volumes. Additionally, we also provide a new set of augmentation and domain-randomization techniques based on the use of spherical harmonics, to synthesize cortical sections of ex-vivo human brains acquired using LSM. This is particularly helpful in enabling zero-shot segmentation, on previously unseen data.

Installation

Clone the repository

git clone https://github.com/lincbrain/lumin
cd lumin

Create new conda environment using the environment.ymlfile:

conda env create --name lsm --file=environment.yml

Next, install the code as an editable package in the conda environment created above.

pip install -e .

Usage

We provide 3 basic functions in the repository, which are described below.

1. Distributed Segmentation
2. Light-Sheet Microscopy Synthesis
3. Analysis

The method to use the code for any of the above functions is more or less the same, i.e.: define a configuration file for each of its corresponding scripts.

Distributed Segmentation

To run distributed segmentation, one can invoke the distributed_segment.pyscript as demonstratated below.

    python distributed_segment.py --config ./configs/segment/<config>.yaml

Note that the path specified to the config file argument is just an example, and can be changed as per convenience.

Light-Sheet Microscopy Synthesis

[TODO]

Analysis

One can perform post-hoc, heuristic analyses on the segmentation and distrbuted stitching algorithms by running the following

python analysis.py --config ./configs/analysis/<config>.yaml

We provide 2 example config files in ./configs/analysis/, where we also briefly describe the function of each of the parameters used.