Brain Stroke Detector

Objective

This repository contains the capstone project for the ML Zoomcamp course provided by DataTalks.Club. The brain stroke detector is currently available online at http://stroke.peco602.com/.

Context

The objective of this project is to develop a stroke detector for brain CT scans. Stroke is the second leading cause of death and the major cause of disability in the world, with an annual mortality of 5.5 million [ref]. In 2017, the stroke-related costs was 45.5 billion USD in USA [ref] and 60 billion Eur in Europe, with health care accounting for 27 billion EUR (45%), representing 1.7% of health expenditure [ref].

Dataset

The CT scan image dataset can be downloaded from Kaggle at this link and contains both brains affected by a stroke and healthy ones.

Deployment

The service is dockerised and can be easily deployed via the following steps:

1. Clone the brain-stroke-detector repository locally:

   $ git clone https://github.com/Peco602/brain-stroke-detector.git

2. Install the pre-requisites necessary to run the service:

   $ cd brain-stroke-detector
   $ sudo apt install make
   $ make prerequisites

   It is also suggested to add the current user to the docker group to avoid running the next steps as sudo:

   $ sudo groupadd docker
   $ sudo usermod -aG docker $USER

   then, logout and log back in so that the group membership is re-evaluated.

3. Pull the Docker images:

   $ make pull

4. Launch the service:

   $ make run

Once ready, the following services will be available:

Service	Port	Interface	Description
Flask	8080	0.0.0.0	API service
Streamlit	8051	0.0.0.0	Web application

Testing

The stroke detector can be accessed via the Streamlit web application available at http://localhost:8051 or by sending the CT scans to be tested directly to the service APIs via curl:

$ curl -F "img=@./data/normal.jpg" localhost:8080/predict
{"stroke": false}
$ curl -F "img=@./data/stroke.jpg" localhost:8080/predict
{"stroke": true}

Disposal

The service can be disposed via:

$ make kill

Training

The Jupyter notebook notebook.ipynb contains the model experiments. A Convolutional Neural Network (CNN) is used to perform stroke detection on the CT scan image dataset. Since the dataset is small, the training of the entire neural network would not provide good results so the concept of Transfer Learning is used to train the model to get more accurate results. Transfer Learning represents a method wherein a model developed for a particular task is used as a starting point for another task. In particular, the InceptionV3 model with imagenet weights will be used for the current task.

The training can be performed again via the following steps:

1. Download the stroke dataset from here

2. Extract the dataset in the data folder with the following structure:

   data
       Brain_Data_Organised
           Normal
           Stroke
   

3. Configure the development evironment:

   $ make setup

4. Launch the training script:

   $ make train

5. Re-build the container with the re-trained model:

   $ make build

6. Launch the service:

   $ make run
   

GitHub Actions

• Continuous Integration: On every push and pull request on main and dev branches, the Docker images are built, tested and then pushed to DockerHub.

Applied technologies

Name	Scope
Jupyter Notebooks	Exploratory data analysis and selection of the best model.
Docker	Application containerization.
Docker-Compose	Multi-container Docker applications definition and running.
Flask	Web server.
pytest	Python unit testing suite.
pylint	Python static code analysis.
black	Python code formatting.
isort	Python import sorting.
Pre-Commit Hooks	Simple code issue identification before submission.
GitHub Actions	CI/CD pipelines.
Streamlit	Web application

Disclaimer

This prediction service has been developed as the capstone project for the ML Zoomcamp course from DataTalks.Club. It does not provide medical advice and it is intended for informational purposes only. It cannot be considered a substitute for professional medical advice, diagnosis or treatment. Never ignore professional medical advice in seeking treatment because of something you have read here.