Installation guide

1. Clone git repository

$ cd /path/to/desired/location
$ git init
$ git clone https://github.com/grace00821/zebrafish_pipeline.git

2. Create conda environment

Mac

$ conda env create --file=mac.yml -n <ENV_NAME> python=3.11

Windows

$ conda env create --file=windows.yml -n <ENV_NAME>

Linux

$ conda env create -f linux.yml -n <ENV_NAME>

3. Activate conda environment

$ conda init
$ conda activate ENV_NAME

Quick start

Locate the working directory to repository $ cd /path/to/repository

For CPU usage

$ python main.py /path/to/ndpi/file -1

For GPU usage

Replace the value 0 with any GPU device number you would like to use.

$ python main.py /path/to/ndpi/file 0

Output of the analysis (blood counts) will be output into the output folder as a .txt file

Pipeline details

This pipeline serves to count zebrafish (Danio rerio) blood cells from digital blood smear slide images of ndpi

Cell recognition and classification (model/detection.py)

Dataset

YOLOv11 model was trained on.

Train: 158 (15023 cells) Test: 38 (5666 cells)

Model performance

Training dataset

Last update: 8 Jan 2025

Author: Eunhye Yang

Description:

After meeting on 8 Jan 2025, refinement of qc step required
Previous version of pipeline: QC on each image only once
Current version of pipeline: if cluster, cut image into half and retry QC
If smaller image standard, run through YOLO
If smaller image still cluter, cut the samller image into half again
Each image should result in 4 smaller images at most to minimize time and computation resources