TumorNet is a web application designed for breast tumor classification, leveraging advanced Artificial Intelligence techniques. The application was developed as part of the thesis titled, "Design and Implementation of a Web Application for Breast Tumor Classification through Convolutional Neural Networks" and further enriched by the accompanying research paper, "Evaluating Deep Learning Architectures for Breast Tumor Classification and Ultrasound Image Detection Using Transfer Learning." The primary objective of this project was to create an open-source web application capable of classifying breast ultrasound images. The system first determines whether an uploaded image is relevant for analysis, then identifies whether the image depicts normal or tumorous breast tissue. For images classified as tumorous, the system further distinguishes between benign and malignant tumors. This multi-step classification process is powered by a sequence of pre-trained deep learning models that were fine-tuned using comprehensive datasets.
The research focused on evaluating the effectiveness of binary classification systems over traditional three-class classification approaches. Although the results show that breaking the task into two stages—first distinguishing normal tissue from tumorous tissue, and then categorizing benign and malignant tumors—did not significantly improve accuracy, the sensitivity and F1 score metrics suggest that this architecture is more effective for the task. This approach has the potential to improve diagnostic accuracy and efficiency in clinical settings, enhancing the early detection of breast cancer and providing valuable support to medical professionals in delivering better patient care. The architecture of TumorNet, showcasing its backend workflow and system components, is illustrated in the image below:
The live demonstration of the application can be accessed here
The current version of the application is v2.1. It has been optimized for minimal latency and fastest response times, with the application hosted on dedicated servers at the TelSiP Research Laboratory, University of West Attica, Athens, Greece.
The software environment for model development was built on Python 3.10.12, utilizing TensorFlow and Keras as the primary frameworks for machine learning and deep learning. To support these frameworks, several auxiliary libraries were included: scikit-learn, Matplotlib, Seaborn, OpenCV, and Pillow. These libraries provided the necessary tools for data preprocessing, model evaluation, and visualization, ensuring a comprehensive and efficient workflow throughout the model development and experimentation process.
The server was developed using Flask, a lightweight Python web framework, which facilitated API creation and backend logic. Additional server-side dependencies and libraries are detailed in the requirements.txt file for reproducibility and setup consistency.
The frontend interface was developed with React, a modern JavaScript library for building user interfaces. For interactive and dynamic graph rendering, various React-compatible visualization libraries were integrated, ensuring a responsive and user-friendly data presentation.
The communication between the frontend and backend of the application is streamlined through a single endpoint, ensuring efficient and secure data exchange.
When the user selects an image for analysis The frontend sends the image as a base64 encoded message to the backend using a POST request.
const handlePrediction = () => {
setIsPredicting(true); // Set predicting state to true
let message = {
image: base64Image
};
// Send data to server. Based on the port running the predict_app.py (Here is local host port 5000 /predict)
fetch("http://127.0.0.1:5000/predict", {
method: 'POST',
body: JSON.stringify(message),
})
.then(response => response.json())
.then(data => {
setTimeout(() => {
setPrediction(data); // Update prediction state to true
console.log(data); //Print the prediction. Debugging purposes only
setIsPredicting(false); // Update prediction state to false
}, 3000);
})
.catch(error => {
console.error('Error:', error);
setIsPredicting(false); //If an error occurs set predicting state to false
});
};Upon receiving the request, the backend decodes the message, preprocesses the image to prepare it for analysis, passes it through the models to generate a prediction, and constructs a structured response object as shown below.
response = {
'irrelevant': prediction1_list,
'tumorous': prediction2_list,
'malignant': prediction3_list
}As depicted in the diagram below, Nginx serves as an intermediary, routing requests between the frontend and backend containers to ensure secure and controlled communication. It enforces security by preventing unauthorized external access, ensuring that only the frontend can communicate with the backend.
The communication flow is managed within a Docker Compose setup, ensuring secure and efficient interaction between the frontend and backend components.
Before cloning the GitHub repository, ensure that the following are installed on your machine:
Git: For cloning the repository. Python 3.10.12 (or later): To run the project.
Clone the repository by running the following command:
git clone https://github.com/christopherkormpos/TumorNet-thesis.gitNavigate to the backend directory and install the required Python dependencies by executing:
pip install -r requirements.txtAfter the dependencies are installed, start the backend by running:
python predict_app.pyThe backend server will be accessible at localhost:5000
To start the frontend, which runs on port 3000, navigate to the frontend directory and execute:
npm install
npm startThe frontend will be accessible at localhost:3000
I would like to express my sincere gratitude to Grigorios Koulouras for his invaluable guidance, mentorship, and unwavering support as both my supervisor and co-writer of the paper; to PhD candidates Fotios Zantalis and Stylianos Katsoulis for their collaboration as co-writers, whose insightful contributions significantly enriched the research.
If you encounter any issues or bugs with the application, or if you face difficulties while building the project locally, feel free to reach out to me:
- Email: christopher.kormpos@gmail.com and ckormpos@uniwa.gr
- GitHub: https://github.com/christopherkormpos
- LinkedIn: LinkedIn Profile
This project is licensed under the MIT License - see the LICENSE file for details.