Brain Tumor Classification & Segmentation 🧠

Project Overview

This project implements a complete computer vision pipeline for brain tumor classification and segmentation using MRI images.

• Classification: Detect tumor presence using ResNet18.
• Segmentation: Extract tumor regions using ResUNet with a ResNet18 encoder.
• Deployment: End-to-end system ready for demo or further evaluation.
• Role: Model training, preprocessing, and deployment.
• Achievement: Finalist in an AI Hackathon.

🔧 Technologies & Tools

• Programming Languages: Python
• Libraries: PyTorch, torchvision, segmentation_models_pytorch, scikit-learn, numpy, pandas, matplotlib, PIL
• Deployment: Streamlit
• Version Control: Git, GitHub

🧩 Pipeline Overview

1. Dataset & Loading

• MRI images and corresponding tumor masks.
• Split into train (80%), validation (20%), and test sets.

2. Preprocessing & Augmentation

• Resize images/masks to 224×224 (ResNet18) or 256×256 (ResUNet).
• Data augmentation:
  • Horizontal/vertical flips
  • Random rotations
  • Color jitter (classification only)
• Convert to tensor; round masks to 0/1.

3. Model Architecture

• ResNet18: Classifier for tumor detection (4 classes).
• ResUNet: Segmentation model with ResNet18 encoder, 1-channel output mask.

4. Training

• Loss Functions:
  • Classification: CrossEntropyLoss
  • Segmentation: DiceLoss (binary)
• Optimizer: Adam, lr=1e-4
• Scheduler: ReduceLROnPlateau (classification)
• Early stopping and model checkpointing based on validation performance.

5. Evaluation & Inference

• Predict tumor class or segmentation masks for test images.
• MC Dropout implemented for uncertainty estimation.
• Segmentation outputs thresholded at 0.5.

6. Results

ResNet18 Evaluation

              precision    recall  f1-score   support
glioma           0.99       0.98      0.99       254
meningioma       0.98       0.98      0.98       306
no_tumor         0.99       1.00      0.99       140
pituitary        0.99       0.99      0.99       300

accuracy                               0.99      1000
macro avg        0.99       0.99      0.99      1000
weighted avg     0.99       0.99      0.99      1000

6. Deployment

• Save best models: best_resnet18.pth, best_resunet.pth.
• Streamlit app allows users to upload images and get predicted class and tumor mask.

Demo Images

Classification and Segmentation Results

LLM-generated Medical Report

📦 Environment Setup Guide

1. Install Anaconda

Anaconda is recommended for managing Python environments and packages.

• Step 1: Download Anaconda:
  👉 https://www.anaconda.com/download
• Step 2: Follow installation instructions.
  Open Anaconda Prompt (Windows) or terminal (macOS/Linux) after installation.

2. Create the Conda Environment

• Step 1: Clone this repository

  git clone https://github.com/<your-username>/<your-repo-name>.git
  cd <your-repo-name>
  

• Step 2: Create the environment from environment.yml:

  conda env create -f environment.yml
  

• Step 3: Activate the environment:

  conda activate <env-name>
  

🚀 Running the Streamlit App

###1. Navigate to the project root directory

Ensure you are in the parent directory of the App/ folder:

project-root/
│
├── App/
│   └── Streamlit_Demo/
│       └── app.py
├── environment.yml
└── README.md

2. Run the app

streamlit run App/Streamlit_Demo/app.py

Terminal will show:

You can now view your Streamlit app in your browser.

  Local URL: http://localhost:8501
  Network URL: http://192.168.x.x:8501

Open the Local URL in your browser to access the app.

Stop anytime with Ctrl + C.