Brain Tumor Classification using Deep Learning

---

---

---

Table of Contents

• Executive Summary
  • Problem Statement
  • Solution
  • Key Results
• Overview
  • Motivation
  • Scientific Significance
  • Key Capabilities
  • Use Cases
• Research Background
  • Peer-Reviewed Research
  • Related Work
• Technology Stack
• Architecture
  • Model Architecture
  • Technical Components
• Methodology
  • Data Processing Workflow
• Performance Metrics
• Installation & Quick Start
  • Requirements
  • Installation
  • Usage
• Expected Outcomes
• Potential Applications
• Milestones
• Contributing
• License
• Contact

---

Executive Summary

Problem Statement

Brain tumor detection and classification remains one of the most challenging areas in medical imaging. Traditional diagnostic methods suffer from:

• Time-consuming processes requiring extensive manual analysis
• Subjectivity in interpretation leading to inconsistent diagnoses
• Human errors due to fatigue and cognitive limitations
• Limited scalability to handle increasing patient volumes

Solution

This project implements an automated brain tumor classification system using deep learning with PyTorch. The solution leverages Convolutional Neural Networks (CNNs) to analyze MRI scans and classify tumor types with high accuracy, providing rapid and objective diagnostic support.

Key Results

• Automated classification of multiple tumor types (glioma, meningioma, pituitary tumor)
• Reduced diagnostic time from hours to seconds
• High accuracy achieved through state-of-the-art CNN architecture
• Scalable solution capable of processing large volumes of medical images

---

Overview

Motivation

Brain tumor detection and classification is one of the most challenging areas in medical imaging. Traditional methods often suffer from:

• Time-consuming processes
• Subjectivity in diagnosis
• Human errors

This project leverages Deep Learning techniques to provide a solution that automates the classification of brain tumors from MRI scans. The goal is to enhance diagnostic efficiency and accuracy.

Scientific Significance

Using state-of-the-art Deep Learning models, this project aims to:

• Automate medical image analysis
• Classify tumor types with high precision
• Improve diagnostic speed and reduce human intervention

Key Capabilities

• Multi-class tumor classification: Distinguish between glioma, meningioma, pituitary tumor, and non-tumor cases
• Real-time inference: Process MRI scans in seconds
• Transfer learning support: Fine-tune on custom datasets
• Explainable AI: Visualize model decision-making through activation maps
• Clinical integration ready: REST API for PACS system integration

Use Cases

• Neuro-oncology clinics: Assist oncologists in tumor detection and treatment planning
• Radiology departments: Enhance radiologist workflow with automated pre-screening
• Telemedicine: Enable remote diagnosis in underserved areas
• Medical research: Facilitate large-scale epidemiological studies
• Educational tools: Training platform for medical students

---

Research Background

Peer-Reviewed Research

This project builds upon established research in medical image analysis using deep learning. The approach is informed by several key studies:

1. CNN-Based Brain Tumor Classification

Reference: https://arxiv.org/abs/1802.10200

Afshar, P., Mohammadi, A., & Plataniotis, K. N. (2018). "Brain Tumor Type Classification via Capsule Networks." 25th IEEE International Conference on Image Processing (ICIP).

Approach Summary:

• Method: Used Capsule Networks (CapsNets) for hierarchical feature learning
• Dataset: 3,064 T1-weighted contrast-enhanced MRI images (3 tumor types)
• Key Innovation: Spatial relationship preservation through dynamic routing
• Results: Achieved 86.56% accuracy, outperforming traditional CNNs
• Relevance: Demonstrated effectiveness of deep learning for multi-class tumor classification

2. Transfer Learning for Medical Imaging

Reference: https://www.sciencedirect.com/science/article/abs/pii/S0010482519302148

Deepak, S., & Ameer, P. M. (2019). "Brain tumor classification using deep CNN features via transfer learning." Computers in Biology and Medicine, 111, 103345.

Approach Summary:

• Method: Transfer learning using pre-trained GoogleNet with SVM classifier
• Dataset: 3,064 brain MRI images across 3 tumor categories
• Key Innovation: Feature extraction from pre-trained networks + traditional ML classifier
• Results: 97.1% accuracy using 5-fold cross-validation
• Relevance: Validates transfer learning approach for limited medical datasets

Approach Summary:

• Method: Multi-scale CNN architecture with attention mechanisms
• Key Innovation: Attention-guided feature fusion across multiple scales
• Results: Robust performance across varying image qualities
• Relevance: Inspires multi-scale feature extraction strategy

Related Work

The project integrates techniques from:

• Computer vision: ResNet, DenseNet architectures for feature extraction
• Medical imaging: DICOM processing, image normalization techniques
• Machine learning: Cross-validation, hyperparameter optimization
• Clinical workflow: Integration with existing hospital information systems

---

Technology Stack

Core Technologies

Libraries & Frameworks

• Deep Learning: PyTorch, TorchVision, ONNX
• Data Processing: NumPy, Pandas, PIL, OpenCV
• Visualization: Matplotlib, Seaborn, TensorBoard
• Model Evaluation: scikit-learn, torchmetrics
• Medical Imaging: SimpleITK, NiBabel

Development Tools

• Version Control: Git, GitHub
• Containerization: Docker
• Experiment Tracking: Weights & Biases / MLflow
• Code Quality: Black, Flake8, pytest

---

Architecture

Model Architecture

Framework: PyTorch - A leading framework for deep learning.

Model: Convolutional Neural Network (CNN) - A powerful architecture for image analysis consisting of:

• Convolutional layers for automatic feature extraction
• Batch normalization for training stability
• Max pooling for dimensionality reduction
• Fully connected layers for classification
• Dropout for regularization

Image Type: MRI Scans - T1-weighted contrast-enhanced brain MRI images

Technical Components

• Dataset Splitting: The dataset is split into training (70%), validation (15%), and test (15%) sets for robust model evaluation
• Convolution Output Dimension Calculation: Dynamically compute the output dimensions of convolution layers
• Max Pooling: Use max pooling to reduce dimensionality while preserving critical features
• Feature Extraction: Convolutional layers automatically extract relevant features from raw images
• Data Augmentation: Random rotations, flips, zooms, and intensity variations

---

Methodology

---

---

Data Processing Workflow

1. Data Exploration: Conducting in-depth analysis to understand data patterns, missing values, and distribution
2. Preprocessing: Image normalization, resizing to 224×224, and intensity standardization
3. Image Augmentation: Applying transformations like rotations, flips, and zoom to increase dataset diversity and improve generalization
4. Model Training: Training the CNN using a supervised learning approach on labeled MRI data with Adam optimizer
5. Multi-Class Classification: The model classifies MRI scans into different categories based on tumor types
6. Validation: Continuous monitoring of validation metrics to prevent overfitting
7. Evaluation: Comprehensive testing on held-out test set using multiple metrics

---

Performance Metrics

---

---

To evaluate the performance of the model, the following metrics are used:

• Accuracy: Overall correctness of the model's predictions
• Precision: The proportion of true positive predictions
• Recall: The proportion of true positives among all actual positive instances
• F1 Score: A balanced metric combining precision and recall
• Confusion Matrix: Detailed breakdown of classifications per tumor type
• ROC-AUC: Area under the receiver operating characteristic curve

---

Installation & Quick Start

Requirements

To run this project, the following packages are required:

• Python 3.8+
• PyTorch 2.0+ (Deep learning framework)
• NumPy (Numerical operations)
• Pandas (Data handling)
• Matplotlib (Data visualization)
• scikit-learn (For model evaluation)
• torchvision (Image transformations)
• Pillow (Image processing)

Installation

Clone the repository and install dependencies:

git clone https://github.com/dhou22/Brain-Cancer-Detection-CNN.git
cd Brain-Cancer-Detection-CNN
pip install -r requirements.txt

Usage

Training the Model

Train the model using the following command:

python train.py --epochs 50 --batch-size 32 --lr 0.001

Performing Inference

To classify a new MRI scan, use the following Python script:

from model import CNN_TUMOR
import torch
from torchvision import transforms
from PIL import Image

# Load the trained model
model = CNN_TUMOR(params)
model.load_state_dict(torch.load("model_weights.pth"))
model.eval()

# Prepare the image for prediction
img = Image.open("path_to_image.jpg")
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], 
                        std=[0.229, 0.224, 0.225])
])
img_tensor = transform(img).unsqueeze(0)

# Predict the tumor type
with torch.no_grad():
    output = model(img_tensor)
    prediction = torch.argmax(output, dim=1)
    
tumor_types = ['Glioma', 'Meningioma', 'Pituitary', 'No Tumor']
print(f"Predicted tumor type: {tumor_types[prediction.item()]}")

---

Expected Outcomes

• Improved diagnostic accuracy: Reducing human error in detecting brain tumors
• Faster medical image processing: Speeding up the diagnosis and treatment planning
• Reduced human bias: Delivering consistent results in tumor classification
• Support for healthcare professionals: Assisting clinicians in accurate decision-making
• Cost reduction: Decreasing diagnostic costs through automation

---

Potential Applications

• Neuro-oncology: Use in clinical settings to assist oncologists in tumor detection and treatment planning
• Radiology departments: Enhancing the workflow of radiologists by automating classification
• Clinical Decision Support Systems: AI-powered tools that provide support for clinicians in diagnosing brain tumors
• Medical Research: Facilitating further research on brain tumor types and AI models for medical imaging
• Telemedicine: Enabling remote diagnosis in areas with limited access to specialists
• Drug development: Supporting pharmaceutical research through automated tumor characterization

---

Milestones

• ✅ Phase 1: Data collection and preprocessing (Completed)
• 🔄 Phase 2: Model training and optimization (In progress)
• 📅 Phase 3: Evaluation and real-world testing (Upcoming)
• 📅 Phase 4: Clinical validation and deployment (Future)

---

Contributing

We welcome contributions! If you'd like to help with the project, please:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Please ensure your code follows PEP 8 style guidelines and includes appropriate tests.

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Contact

For questions or feedback, reach out to:

• Email: dhouha.meliane@esprit.tn
• LinkedIn: https://www.linkedin.com/in/dhouha-meliane/
• GitHub: @dhou22

---

References

1. Afshar, P., Mohammadi, A., & Plataniotis, K. N. (2018). Brain Tumor Type Classification via Capsule Networks. 25th IEEE International Conference on Image Processing (ICIP), 3129-3133.

2. Deepak, S., & Ameer, P. M. (2019). Brain tumor classification using deep CNN features via transfer learning. Computers in Biology and Medicine, 111, 103345.

---

Disclaimer: This project is intended as a research tool and should not replace professional medical diagnoses. Always consult healthcare professionals for accurate diagnoses. This system is designed to assist, not replace, clinical decision-making.

---

Acknowledgments: This research was inspired by and builds upon the pioneering work of researchers in medical AI and deep learning for healthcare applications.