ancer Detection Framework

Advanced deep learning framework for automated detection of malignant and benign breast tumors from histopathological images.

Overview

This framework implements state-of-the-art computer vision techniques for binary classification of breast cancer histology samples. The system processes microscopic tissue images captured at multiple magnification levels (40X-400X) to distinguish between benign and malignant cases.

Dataset Information

Utilizes the BreakHis dataset containing 7,909 microscopic images from 82 patients:

• Benign samples: 2,480 images (adenosis, fibroadenoma, phyllodes tumor, tubular adenoma)
• Malignant samples: 5,429 images (ductal carcinoma, lobular carcinoma, mucinous carcinoma, papillary carcinoma)
• Resolution: 700x460 pixels, RGB, PNG format

Architecture

The framework employs a modular design pattern with clear separation of concerns:

├── core/
│   ├── engine.py          # Training orchestration
│   ├── evaluator.py       # Model assessment
│   └── predictor.py       # Inference pipeline
├── infrastructure/
│   ├── datapipeline.py    # Data loading and preprocessing
│   └── checkpointing.py   # Model persistence
├── networks/
│   ├── architectures.py   # Neural network definitions
│   └── objectives.py      # Loss functions
├── monitoring/
│   ├── metrics.py         # Performance tracking
│   └── visualization.py   # TensorBoard integration
├── configuration/
│   └── settings.py        # Hyperparameters
└── main.py                # Entry point

Installation

python -m venv venv_medical
source venv_medical/bin/activate  # Windows: venv_medical\Scripts\activate
pip install torch torchvision numpy pillow pyyaml tensorboard scikit-learn

Usage

Training

python main.py --mode train --config configuration/experiment.yaml

Evaluation

python main.py --mode evaluate --checkpoint outputs/models/best.pth

Inference

python main.py --mode predict --image path/to/image.png --checkpoint outputs/models/best.pth

Configuration

Modify configuration/experiment.yaml to adjust:

• Network architecture (DenseNet121, ResNet50, EfficientNet)
• Batch size and learning rate
• Augmentation strategies
• Training duration

Performance Metrics

The framework tracks:

• Classification accuracy
• Precision, recall, F1-score
• ROC-AUC
• Confusion matrix

Results are logged to TensorBoard:

tensorboard --logdir outputs/logs

Dataset Preparation

1. Download BreakHis from Kaggle
2. Extract to datasets/breakhis/
3. Structure should be:

datasets/breakhis/
├── benign/
│   ├── adenosis/
│   ├── fibroadenoma/
│   └── ...
└── malignant/
    ├── ductal_carcinoma/
    ├── lobular_carcinoma/
    └── ...

Citation

If you use this framework, please cite:

@article{breakhis2024,
  title={Histopathology Cancer Detection Framework},
  author={Medical Imaging Research Lab},
  year={2024}
}

License

Apache License 2.0 - See LICENSE file for details