nuclei-segmentation

Development of machine learning model for instance segmentation of nuclei cells for Kaggle Data Science Bowl 2018 challenge.

Installation

First, clone the GitHub repository:

git clone https://github.com/yelmokht/nuclei-segmentation.git

Make sure that poetry is installed:

pip install poetry

Then, install all necessary packages:

poetry install

Note: If you are on Windows and use an Intel processor with an iGPU, you may encounter this error: ModuleNotFoundError: No module named 'tensorflow'. To resolve this, do these commands on the terminal in order to install tensorflow-intel 2.15.0 in your virtual environment:

poetry shell
pip install tensorflow-intel==2.15.0

Usage

Once installation is complete, you can run the application:

poetry run python src/main.py

Data

This application uses the dataset from Kaggle Data Science Bowl 2018 available here. If you use the application, the dataset will automatically be downloaded in the data folder.

data
    ├── dsb-2018
    ├── unzipped
    └── data-science-bowl-2018.zip

Models

This application utilizes deep learning and convolutionnal neural networks (CNN) with tensorflow/keras library. A pretrained model for this challenge is accessible here. If you use the application, the model will automatically be downloaded in the models folder. You can also train a model inside the application using Train a new model feature but a GPU is recommended to function properly. Another option is to use Google Colab to train a model: you can use nuclei.ipynb available in the notebook folder. After training, the model and its history should normally be saved in your Google Drive. You can then insert them into the models folder following this structure:

models
    ├── model_name
    │   ├── model.h5
    │   └── history.csv

Screenshots of application