Hemovision

Automated blood cells classification for diagnosis and research

Presentation and Installation

This repository contains the code for our project Blood cells classification, developed during our Data Scientist training at DataScientest.

The goal of this project is to to identify the different types of blood cells using computer vision algorithms. The density and relative abundance of blood cells in the smear is crucial for the diagnosis of many pathologies, such as leukemia, which relies on the ratio of lymphocytes. The identification of abnormal leukocytes in pathologies such as leukemia could complete this first part. Developing a tool capable of analyzing cells from blood smears could facilitate the diagnosis of certain pathologies, but could also be used for research purposes.

This project was developed by the following team :

• Bertrand-Elie DURAN (GitHub)
• Sergey SASNOUSKI (GitHub / LinkedIn)
• Joseph LIEBER (GitHub / LinkedIn)
• Kipédène COULIBALY (GitHub / LinkedIn)

You can browse and run the notebooks.

You will need to install the dependencies (in a dedicated environment) :

pip install -r requirements.txt

Hemovision App

• Install streamlit.
• On the "predictions" page, choose your model (8 classes or 11 classes, see the modeling report for details).
• Import your own image or choose from the images provided.
  NB: the result may take up to 8 or 9 seconds to be displayed.

To run the app (be careful with the paths of the files in the app):

conda create --name my-awesome-streamlit python=3.9
conda activate my-awesome-streamlit
pip install -r requirements.txt
streamlit run app.py

The app should then be available at localhost:8501.