🏥 Breast Cancer Detection System

A comprehensive machine learning solution for breast cancer detection with a user-friendly web interface. This project compares multiple state-of-the-art algorithms (Random Forest, XGBoost, CatBoost, LightGBM, SVM, and Neural Networks) and provides real-time predictions through an intuitive web application.

---

📋 Table of Contents

• Features
• Project Structure
• Installation
• Quick Start
• Usage Guide
• Web Interface
• API Documentation
• Model Comparison
• Understanding the Results
• Deployment
• Dataset Information
• Medical Disclaimer
• Contributing
• License

---

Interactive Interface demonstrating Probability Scoring and Feature Explanation

---

✨ Features

🎯 Core Features

• 📊 Multiple Model Comparison: Evaluates 6 different machine learning algorithms

  • Random Forest
  • XGBoost
  • CatBoost
  • LightGBM
  • Support Vector Machine (SVM)
  • Neural Network

• 🎯 Automatic Best Model Selection: Selects the best performing model based on F1-score

• 🚀 Production-Ready API: FastAPI-based REST API for real-time predictions

• 💻 User-Friendly Web Interface:

  • Beautiful, responsive web UI
  • Interactive form with all 30 features
  • Real-time predictions
  • Color-coded results (Green for Benign, Red for Malignant)

• 📚 Educational Tooltips:

  • Help icons (?) next to each field
  • Plain-language explanations for all medical terminology
  • Tooltips on hover with detailed descriptions
  • Clear explanations of Malignant vs Benign

• 🎲 Random Example Generation:

  • Load random examples from the dataset
  • Different example each time
  • Pre-filled form for quick testing

• 📈 Comprehensive Evaluation:

  • Accuracy, Precision, Recall, F1-score
  • 5-fold Cross-Validation
  • Detailed classification reports
  • Confusion matrix analysis

• 🔧 Deployment Ready:

  • Docker support
  • Cloud deployment guides (Render, Heroku, Railway, AWS, GCP, Azure)
  • Traditional hosting deployment instructions

---

📁 Project Structure

Breast Cancer Detection/
├── Breast_cancer_dataset.csv      # Dataset with 569 samples and 30 features
├── train_models.py                # Model training and comparison script
├── app.py                         # FastAPI application with web interface
├── requirements.txt               # Python dependencies
├── static/
│   └── index.html                # User-friendly web interface
├── Dockerfile                     # Docker container configuration
├── docker-compose.yml             # Docker Compose setup
├── Procfile                       # Heroku deployment configuration
├── render.yaml                    # Render.com deployment configuration
├── runtime.txt                    # Python version specification
├── deploy.sh                      # Quick deployment script
├── setup_hosting.sh               # Hosting deployment script
├── start_server.sh                # Server start script
├── passenger_wsgi.py              # WSGI entry point for hosting providers
├── DEPLOYMENT.md                  # Comprehensive deployment guide
├── DEPLOYMENT_QUICKSTART.md       # Quick deployment guide
├── DEPLOYMENT_HOSTING.md          # Traditional hosting deployment guide
├── RENDER_DEPLOYMENT.md           # Render.com specific guide
├── .gitignore                     # Git ignore file
└── README.md                      # This file

---

🚀 Installation

Prerequisites

• Python 3.8 or higher
• pip (Python package manager)

Step 1: Clone the Repository

git clone https://github.com/ShaonINT/Breast_Cancer_Detection.git
cd Breast_Cancer_Detection

Step 2: Install Dependencies

pip install -r requirements.txt

Required packages:

• pandas
• numpy
• scikit-learn
• xgboost
• catboost
• lightgbm
• fastapi
• uvicorn
• pydantic
• joblib

---

⚡ Quick Start

1. Train the Models

python train_models.py

This will:

• Load and preprocess the dataset
• Train all 6 models
• Evaluate each model
• Select the best model (highest F1-score)
• Save model files (best_model.pkl, model_metadata.pkl)

Expected Output:

================================================================================
BREAST CANCER DETECTION - MODEL COMPARISON
================================================================================

Training and evaluating models...

Random Forest:    Accuracy: 0.9649, F1-Score: 0.9500
XGBoost:          Accuracy: 0.9737, F1-Score: 0.9630
CatBoost:         Accuracy: 0.9561, F1-Score: 0.9367
LightGBM:         Accuracy: 0.9649, F1-Score: 0.9500
SVM:              Accuracy: 0.9737, F1-Score: 0.9630
Neural Network:   Accuracy: 0.9737, F1-Score: 0.9630

Best Model: XGBoost (F1-Score: 0.9630)

2. Start the Web Application

python app.py

Then open your browser:

• Web Interface: http://localhost:8000
• API Documentation: http://localhost:8000/docs
• Alternative Docs: http://localhost:8000/redoc

---

📖 Usage Guide

Using the Web Interface

1. Navigate to http://localhost:8000

2. Read the Information Box at the top to understand:

   • What the measurements mean
   • What Malignant and Benign mean
   • How to use the tool

3. Fill in the Form:

   • Hover over the ? icon next to any field for explanations
   • Or click "Load Random Example" to fill with sample data
   • Enter all 30 feature values

4. Get Prediction:

   • Click "Get Prediction" button
   • View results:
     • B (Benign) in green = Non-cancerous
     • M (Malignant) in red = Cancerous
     • Probability percentage
     • Confidence level (High/Medium/Low)

5. Try Another:

   • Click "Load Random Example" for a different sample
   • Or "Clear Form" to start fresh

Understanding the Form Fields

Each field has:

• Label: The feature name
• ? Icon: Hover for detailed explanation
• Help Text: Brief description below the label
• Input Field: Enter the numeric value

Field Categories:

• Mean Features: Average measurements across all cells (10 fields)
• Standard Error Features: Variation/uncertainty in measurements (10 fields)
• Worst Features: Most abnormal/severe measurements (10 fields)

---

🌐 Web Interface Features

🎨 User-Friendly Design

• Responsive Layout: Works on desktop, tablet, and mobile
• Color-Coded Results:
  • Green background for Benign (B) results
  • Red background for Malignant (M) results
• Clear Visual Hierarchy: Organized sections for easy navigation

📚 Educational Features

• Interactive Tooltips: Hover over ? icons for explanations
• Plain Language: All medical terms explained in simple language
• Info Box: Overview of measurements and terminology at the top
• Result Explanations: Each prediction includes explanation of what it means

🎲 Example Features

• Random Examples: Get different examples each time
• Real Data: Examples come from actual dataset
• Pre-filled Form: One-click to populate all fields

---

🔌 API Documentation

Base URL

http://localhost:8000

Endpoints

1. GET /

Returns the web interface HTML page.

2. GET /health

Health check endpoint to verify the API and model status.

Response:

{
  "status": "healthy",
  "model_loaded": true,
  "model_name": "XGBoost"
}

3. GET /example

Get a random example from the dataset with proper precision formatting.

Response:

{
  "radius_mean": 17.99,
  "texture_mean": 10.38,
  "perimeter_mean": 122.8,
  ...
}

4. POST /predict

Single prediction endpoint.

Request Body:

{
  "radius_mean": 17.99,
  "texture_mean": 10.38,
  "perimeter_mean": 122.8,
  "area_mean": 1001.0,
  "smoothness_mean": 0.1184,
  "compactness_mean": 0.2776,
  "concavity_mean": 0.3001,
  "concave_points_mean": 0.1471,
  "symmetry_mean": 0.2419,
  "fractal_dimension_mean": 0.07871,
  "radius_se": 1.095,
  "texture_se": 0.9053,
  "perimeter_se": 8.589,
  "area_se": 153.4,
  "smoothness_se": 0.006399,
  "compactness_se": 0.04904,
  "concavity_se": 0.05373,
  "concave_points_se": 0.01587,
  "symmetry_se": 0.03003,
  "fractal_dimension_se": 0.006193,
  "radius_worst": 25.38,
  "texture_worst": 17.33,
  "perimeter_worst": 184.6,
  "area_worst": 2019.0,
  "smoothness_worst": 0.1622,
  "compactness_worst": 0.6656,
  "concavity_worst": 0.7119,
  "concave_points_worst": 0.2654,
  "symmetry_worst": 0.4601,
  "fractal_dimension_worst": 0.1189
}

Response:

{
  "prediction": "Malignant (M)",
  "probability": 0.95,
  "confidence": "High"
}

5. POST /predict/batch

Batch prediction for multiple samples at once.

Request Body:

[
  {
    "radius_mean": 17.99,
    ...
  },
  {
    "radius_mean": 13.54,
    ...
  }
]

Response:

{
  "predictions": [
    {
      "prediction": "Malignant (M)",
      "probability": 0.95,
      "confidence": "High"
    },
    {
      "prediction": "Benign (B)",
      "probability": 0.12,
      "confidence": "High"
    }
  ],
  "count": 2
}

Using the API with Python

import requests

url = "http://localhost:8000/predict"
data = {
    "radius_mean": 17.99,
    "texture_mean": 10.38,
    # ... all other features
}

response = requests.post(url, json=data)
result = response.json()

print(f"Prediction: {result['prediction']}")
print(f"Probability: {result['probability']*100:.2f}%")
print(f"Confidence: {result['confidence']}")

Using the API with cURL

curl -X POST "http://localhost:8000/predict" \
  -H "Content-Type: application/json" \
  -d '{
    "radius_mean": 17.99,
    "texture_mean": 10.38,
    ...
  }'

---

📊 Model Comparison

Supported Models

1. Random Forest: Ensemble learning with decision trees
2. XGBoost: Gradient boosting framework
3. CatBoost: Gradient boosting with categorical features support
4. LightGBM: Fast gradient boosting framework
5. SVM (Support Vector Machine): Kernel-based classification
6. Neural Network: Multi-layer perceptron classifier

Evaluation Metrics

Each model is evaluated using:

• Accuracy: Overall correctness of predictions
• Precision: Correctness of positive (malignant) predictions
• Recall: Ability to find all malignant cases
• F1-Score: Harmonic mean of precision and recall (used for model selection)
• Cross-Validation: 5-fold CV for robust performance estimation

Best Model Selection

The model with the highest F1-score is automatically selected as the best model. F1-score balances precision and recall, making it ideal for medical diagnosis where both false positives and false negatives are important.

---

📖 Understanding the Results

Prediction Output

The model returns one of two predictions:

🟢 Benign (B) - Non-Cancerous

• Meaning: The cells appear normal and healthy
• Result: The tissue is NOT cancerous
• Display: Shown with green color
• What to do: Continue regular monitoring as recommended by your doctor

🔴 Malignant (M) - Cancerous

• Meaning: The cells show abnormalities that may indicate cancer
• Result: The tissue may be cancerous
• Display: Shown with red color
• What to do: Consult a healthcare professional immediately for proper medical evaluation and diagnosis

Confidence Levels

• High: Probability ≥ 80% or ≤ 20% (very confident prediction)
• Medium: Probability between 70-80% or 20-30% (moderately confident)
• Low: Probability between 30-70% (less confident, may require more tests)

Probability

The probability value indicates how confident the model is that the sample is malignant (M). For example:

• 0.95 = 95% probability of being malignant
• 0.12 = 12% probability of being malignant (88% chance it's benign)

---

🚀 Deployment

Quick Deployment Options

Option 1: Render.com (Recommended - Easiest)

1. Go to render.com and sign up
2. Click "New +" → "Web Service"
3. Connect your GitHub repository: ShaonINT/Breast_Cancer_Detection
4. Configure:
   • Build Command: pip install -r requirements.txt
   • Start Command: python app.py
5. Click "Create Web Service"
6. Your app will be live in ~2 minutes!

📖 Detailed Guide: See RENDER_DEPLOYMENT.md

Option 2: Railway.app

1. Go to railway.app
2. Connect GitHub repository
3. Railway auto-detects and deploys
4. Done!

Option 3: Docker

docker build -t breast-cancer-detection .
docker run -p 8000:8000 breast-cancer-detection

Option 4: Traditional Web Hosting

See DEPLOYMENT_HOSTING.md for detailed instructions.

Deployment Guides

• Quick Start: DEPLOYMENT_QUICKSTART.md
• Comprehensive Guide: DEPLOYMENT.md
• Traditional Hosting: DEPLOYMENT_HOSTING.md
• Render.com: RENDER_DEPLOYMENT.md

---

📊 Dataset Information

Dataset Source

Kaggle Dataset: Breast Cancer Dataset

The dataset is available on Kaggle and is based on the Wisconsin Breast Cancer Database (WBCD). This dataset is widely used in machine learning research for breast cancer classification tasks.

Dataset Description

The dataset contains features computed from digitized images of Fine Needle Aspirate (FNA) samples of breast masses. FNA is a diagnostic procedure where a thin needle is used to extract a small sample of cells from a breast mass for microscopic examination.

Key Characteristics:

• Source: Features are derived from images of cell nuclei obtained through FNA procedures
• Purpose: Classify breast masses as benign (non-cancerous) or malignant (cancerous)
• Medical Context: This is a binary classification problem critical for early breast cancer detection
• Data Quality: Well-curated dataset with minimal missing values

Dataset Statistics

• Total Samples: 569 instances
• Features: 30 numerical features
• Target Distribution:
  • B (Benign): 357 cases (62.7%) - Non-cancerous tissue
  • M (Malignant): 212 cases (37.3%) - Cancerous tissue
• Class Imbalance: Slight imbalance toward benign cases (typical in medical datasets)

Feature Categories

The 30 features are organized into three categories, each measuring 10 different characteristics of cell nuclei:

1. Mean Features (10 features): Average measurements across all cells in the image

   • Provides overall characterization of cell nuclei

2. Standard Error Features (10 features): Standard error (variation) in measurements

   • Indicates consistency and variability across cells

3. Worst Features (10 features): Largest (worst/most abnormal) measurements found

   • Captures the most severe abnormalities in cell nuclei

Feature Descriptions

Each of the 10 measured characteristics provides different insights into cell structure:

• Radius: Distance from center to points on the perimeter (size indicator)
• Texture: Standard deviation of gray-scale values (surface appearance variation)
• Perimeter: Distance around the boundary of the cell nucleus
• Area: Size of the cell nucleus (often larger in malignant cells)
• Smoothness: Local variation in radius lengths (boundary smoothness)
• Compactness: Perimeter² / (Area - 1) (how circular/compact the shape is)
• Concavity: Severity of concave portions of the contour (indentations)
• Concave Points: Number of concave portions (frequency of indentations)
• Symmetry: How symmetrical the cell nucleus is (normal cells are more symmetrical)
• Fractal Dimension: Complexity of the boundary ("coastline approximation")

Clinical Significance

This dataset is particularly valuable because:

• Early Detection: Enables classification based on cell characteristics visible in FNA samples
• Non-Invasive: FNA is less invasive than surgical biopsies
• Fast Results: Can provide quicker preliminary diagnosis
• Pattern Recognition: Machine learning models can identify subtle patterns not easily visible to the human eye

Data Preprocessing

In this project, the dataset is preprocessed by:

• Normalizing feature names (removing spaces)
• Encoding target variable (M=1, B=0)
• Handling missing values (if any)
• Splitting into training (80%) and testing (20%) sets with stratification

Note: For detailed explanations of each feature, hover over the ? icons in the web interface!

---

⚠️ Medical Disclaimer

IMPORTANT: This tool is for educational and research purposes only.

• ❌ NOT a substitute for professional medical diagnosis
• ❌ NOT intended for clinical decision-making
• ✅ ALWAYS consult with a qualified healthcare provider
• ✅ Results should be interpreted by medical professionals

This application does not provide medical advice. Always seek the advice of a physician or other qualified health provider with any questions regarding a medical condition.

---

🔧 Technical Details

Model Training Process

1. Data Preprocessing:

   • Remove ID column
   • Normalize feature names (spaces to underscores)
   • Remove unnamed/empty columns
   • Encode target variable (M=1, B=0)
   • Handle missing values

2. Data Splitting:

   • 80% training set
   • 20% test set
   • Stratified split to maintain class distribution

3. Feature Scaling:

   • Automatic scaling for SVM and Neural Network
   • Other models use raw features

4. Model Evaluation:

   • 5-fold cross-validation
   • Multiple metrics (Accuracy, Precision, Recall, F1)
   • Best model selection based on F1-score

Model Files

After training, the following files are generated:

• best_model.pkl: The selected best model (joblib format)
• model_metadata.pkl: Metadata including:
  • Model name
  • Feature names and order
  • Performance metrics
  • Whether scaler is needed
• scaler.pkl: Feature scaler (if required by the model)

---

🛠️ Development

Running Tests

# Test the API
python app.py
# Then visit http://localhost:8000/health

# Test predictions
curl http://localhost:8000/predict -X POST -H "Content-Type: application/json" -d '{...}'

Project Structure Details

• train_models.py: Model comparison and training pipeline
• app.py: FastAPI application with endpoints and model loading
• static/index.html: Complete web interface with explanations
• Deployment files: Docker, Render, Heroku, hosting configurations

---

📝 Usage Examples

Example 1: Web Interface

1. Start the server: python app.py
2. Open browser: http://localhost:8000
3. Click "Load Random Example"
4. Click "Get Prediction"
5. View result: M (Malignant) or B (Benign)

Example 2: API Programmatic Access

import requests

# Get a random example
example = requests.get('http://localhost:8000/example').json()

# Make a prediction
prediction = requests.post(
    'http://localhost:8000/predict',
    json=example
).json()

print(f"Result: {prediction['prediction']}")
print(f"Confidence: {prediction['confidence']}")

Example 3: Batch Predictions

import requests

examples = [
    requests.get('http://localhost:8000/example').json()
    for _ in range(5)
]

results = requests.post(
    'http://localhost:8000/predict/batch',
    json=examples
).json()

for i, result in enumerate(results['predictions']):
    print(f"Sample {i+1}: {result['prediction']}")

---

🎯 Key Features Explained

User-Friendly Terminology

• Mean: Average value across all cells
• SE (Standard Error): How much values vary
• Worst: Most abnormal/severe value
• Benign (B): Non-cancerous (healthy)
• Malignant (M): Cancerous (abnormal)

Interactive Help

Every field has:

• Help icon (?): Hover to see detailed explanation
• Brief description: Plain language summary
• Tooltip: Full explanation of what the measurement means

---

📚 Additional Resources

Deployment Documentation

• DEPLOYMENT_QUICKSTART.md: Quick deployment guide
• DEPLOYMENT.md: Comprehensive deployment options
• DEPLOYMENT_HOSTING.md: Traditional hosting guide
• RENDER_DEPLOYMENT.md: Render.com specific guide

Scripts

• deploy.sh: Quick deployment script
• setup_hosting.sh: Hosting setup automation
• start_server.sh: Server start script

---

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Areas for Improvement

• Additional model architectures
• Enhanced feature engineering
• Better UI/UX improvements
• Documentation improvements
• Performance optimizations

---

📄 License

This project is for educational and research purposes only.

---

👤 Author

Shaon Biswas

• GitHub: @ShaonINT
• Repository: Breast_Cancer_Detection

---

🙏 Acknowledgments

• Dataset Source: Breast Cancer Dataset on Kaggle by wasiqaliyasir
• Original Dataset: Wisconsin Breast Cancer Database (WBCD)
• Libraries: scikit-learn, XGBoost, CatBoost, LightGBM, FastAPI
• Community: Open source machine learning community

---

📞 Support

For issues, questions, or contributions:

• Open an issue on GitHub
• Check the deployment guides
• Review the documentation files

---

⭐ If you find this project useful, please consider giving it a star on GitHub!