Brain Tumor Detection System

A computer vision web application and training pipeline for diagnosing brain tumors from magnetic resonance imaging (MRI) scans.

The system classifies input scans into four diagnostic categories:

• Glioma Tumor
• Meningioma Tumor
• Pituitary Tumor
• No Tumor (Healthy)

---

Repository Structure

• main.py: Flask application server providing image upload endpoints, preprocessing routines, model inference, and template rendering.
• templates/: HTML user interface with responsive CSS for drag-and-drop MRI upload, scan previews, and confidence score displays.
• notebooks/: Jupyter notebooks for exploratory data analysis (EDA), data augmentation experiments, transfer learning benchmarks, and model evaluation.
• models.dvc: Data Version Control (DVC) pointer tracking large trained model checkpoint files without bloating the git history.
• docker/: Dockerfile and deployment configurations for containerized inference.
• pyproject.toml: Modern Python project specification locked to Python 3.12 and managed with UV.
• justfile: Standard task runner exposing environment setup, development server launch, and linting commands.

---

Quick Start Guide

Prerequisites

• UV (Python package manager)
• Just (Command runner)

1. Install Dependencies

Create the Python 3.12 virtual environment and synchronize all dependencies:

just install

2. Launch the Application

Start the Flask web server:

just start

Open your browser and navigate to http://localhost:5000.

3. Development Mode

To run Flask with hot-reloading and debug output enabled:

just dev

---

Model Pipeline & Inference

1. Preprocessing: Uploaded MRI images are decoded via OpenCV, converted to RGB, resized to the network input resolution ($128 \times 128$), and normalized to the range $[0, 1]$.
2. Inference: The processed tensor is passed through the deep convolutional neural network (CNN) feature extractor.
3. Classification: Softmax activation calculates posterior probability distributions across the four target diagnostic classes.
4. Result Presentation: The UI renders the top-predicted class along with percentage confidence levels.

---

Model Tracking with DVC

Model weight files are versioned using DVC to keep the Git repository lightweight. To pull the latest model weights from configured remote storage:

dvc pull

---

Docker Deployment

You can build and run the application inside a container:

# Build the Docker image
docker build -f docker/Dockerfile -t brain-tumor-detection .

# Run the container on port 5000
docker run -p 5000:5000 brain-tumor-detection

---

Code Quality & Maintenance

# Check Python syntax and linting
just lint

# Remove cache files and build artifacts
just clean