micronuclAI

Automated assessment of Chromosomal Instability through quantification of micronuclei (MN) and Nuclear Buds (NBUDs).

micronuclAI is an automated pipeline for reliable quantification of micronuclei of varying size, morphology and location from nuclei-stained images. The pipeline can be divided into 3 steps: 1) segmentation, 2) nuclei isolation, and 3) quantification. In brief, single-cell crops are extracted from high-resolution microscopy images with the help of segmentation masks, which are then used to train a convolutional neural network (CNN) to output the number of MN associated with each cell.

Getting Started

Stream-lit App

A webapp implementation of micronuclAI is available through streamlit under:

GitHub:
https://github.com/SchapiroLabor/micronuclAI_streamlit

Streamlit:
https://micronuclai.streamlit.app

Inference of micronuclei can be achieved in small to medium-sized example images that can be uploaded to the streamlit app. Image data is processed within a virtual machine (VM) on Heicloud, a local Cloud infrastructure provided by University Computing Center Heidelberg, and images are immediately deleted after micronuclei inference. Once micronuclei are inferred, results predictions as well as several plots describing the results are generated and presented to the user within the streamlit app which can also be downloaded as csv files.

Local Installation

Prerequisites

Two input files are required for micronuclAI

1. Nuclei-stained Image
2. Associated Mask

Local Installation

A step by step series of examples that tell you how to get a development environment running locally

Please ensure you have Python>=3.10 installed with

python -version

Pull the repo into your local device

git clone https://github.com/SchapiroLabor/micronuclAI

Install the required libraries

pip install -r requirements.txt

Usage

To use micronuclAI on your own images:

python src/model/prediction2.py -i <path/to/image> -m <path/to/mask> -mod <path/to/model> -o <path/to/output/folder>

A test dataset has been provided in the test_data folder with a mask and its corresponding image file.

python src/model/prediction2.py -i ./test/test_image.tiff -m ./test/test_mask.tiff -mod ./micronuclAI_model/micronuclai.pt -o ./test/output

Parameters and Arguments

Parameter	Short Form	Required	Default	Type	Description
--image	-i	Yes	N/A	String	Pathway to image.
--mask	-m	Yes	N/A	String	Pathway to mask.
--model	-mod	Yes	N/A	String	Pathway to prediction model.
--out	-o	Yes	N/A	String	Path to the output data folder.
--size	-s	No	(256, 256)	List of int	Size of images for training.
--resizing_factor	-rf	No	0.6	Float	Resizing factor for images.
--expansion	-e	No	25	Int	Expansion factor for images.
--precision	-p	No	32	String	Precision for training. Options: ["16-mixed", "bf16-mixed", "16-true", "bf16-true", "32", "64"]
--device	-d	No	"cpu"	String	Device to be used for training.
--batch_size	-bs	No	32	Int	Batch size for training.
--workers	-w	No	0	Int	Number of workers for training/testing.

Citation

https://doi.org/10.1101/2024.05.24.595722

@article{IbarraArellano2024,
  title = {micronuclAI: Automated quantification of micronuclei for assessment of chromosomal instability},
  url = {http://dx.doi.org/10.1101/2024.05.24.595722},
  DOI = {10.1101/2024.05.24.595722},
  publisher = {Cold Spring Harbor Laboratory},
  author = {Ibarra-Arellano,  Miguel A. and Caprio,  Lindsay A. and Hada,  Aroj and Stotzem,  Niklas and Cai,  Luke and Shah,  Shivem and Melms,  Johannes C. and W\"{u}nneman,  Florian and Izar,  Benjamin and Schapiro,  Denis},
  year = {2024},
  month = may 
}

License

micronuclAI offers a dual licensing mode the GNU Affero General Public License v3.0 - see LICENSE and ESSENTIAL_LICENSE_CONDITIONS.txt