Deployed link - 🔗
"That which is measured, improves." – Karl Pearson
This repository demonstrates a Convolutional Neural Network (CNN) model for histopathology image classification, trained on the PatchCamelyon (PCam) dataset. The goal is to classify small image patches of lymph node tissue as tumor (metastatic) or normal.
- PCam is a benchmark dataset for binary classification of histopathology images.
- It contains 327,680 color images (96x96 px) extracted from lymph node scans.
- Each image has a binary label indicating the presence of metastatic tissue.
- Green boxes in example images indicate tumor tissue, which determines a positive label.
Dataset link: PCam on GitHub
PCam provides a challenging benchmark: larger than CIFAR-10 but smaller than ImageNet, and it is trainable on a single GPU.
- Model: CNN-based architecture with
ResNet50as the backbone. - Transfer Learning: Applied to speed up training.
- Interpretability: Grad-CAM heatmaps visualize which regions influence predictions.
Current Implementation Notes:
- For educational purposes, a smaller dataset and fewer epochs were used.
- This results in moderate accuracy, precision, and F1-scores.
- With larger datasets and longer training, the model would likely achieve higher performance.
# Clone the repository
git clone https://github.com/BleeGleeWee/Histopathology-Image-Classification.git
cd Histopathology-Image-Classification
# Create a virtual environment
python -m venv venv
source venv/bin/activate # Linux/Mac
venv\Scripts\activate # Windows
# Install dependencies
pip install -r requirements.txt-
Download PCam
.h5files from PCam dataset. -
Place them inside the
data/folder. -
Open Jupyter Notebook
notebooks/Histopathology_CNN_GradCAM.ipynb. -
Run the notebook step-by-step to:
- Load and preprocess the data
- Train the CNN model
- Evaluate model metrics
- Visualize Grad-CAM overlays
- Use
h5pyorHDF5Matrixfrom Keras to efficiently load large datasets. - Important: Loading the entire dataset in memory may cause MemoryError. Consider streaming batches using generators for large datasets.
Example results using smaller dataset:
| Metric | Class 0 (Normal) | Class 1 (Tumor) | Overall |
|---|---|---|---|
| Precision | 0.65 | 0.82 | 0.74 |
| Recall | 0.88 | 0.54 | 0.71 |
| F1-score | 0.75 | 0.65 | 0.70 |
| Accuracy | - | - | 0.71 |
- Interpretation: Model detects normal patches better than tumor patches.
- Reason: Smaller dataset and fewer epochs limit learning capacity.
- Expectation: Using full PCam dataset and more training epochs should produce better performance.
- Highlights regions in the images that the CNN focuses on.
- Overlayed heatmaps help interpret the model’s decisions.
Example Grid:
Histopathology-Image-Classification/
│
├── data/ # PCam dataset .h5 files
│ ├── camelyonpatch_level_2_split_train_x.h5
│ ├── camelyonpatch_level_2_split_train_y.h5
│ ├── camelyonpatch_level_2_split_valid_x.h5
│ └── camelyonpatch_level_2_split_valid_y.h5
│
├── notebooks/ # Jupyter notebooks
│ └── Histopathology_CNN_GradCAM.ipynb
│
├── src/ # Helper scripts (optional)
│ ├── model.py # CNN/ResNet model building
│ ├── data_loader.py # Functions to load h5 data
│ └── gradcam.py # Grad-CAM functions
│
├── images/ # Example images and Grad-CAM overlays
│
├── requirements.txt # Dependencies
├── README.md
└── .gitignore
- This repository is for educational purposes only.
- Everyone is encouraged to submit pull requests or issues to improve the project.
- Suggestions for optimizing the model, adding more visualizations, or better data handling are welcome.
-
Veeling, B. S., Linmans, J., Winkens, J., Cohen, T., & Welling, M. (2018). Rotation Equivariant CNNs for Digital Pathology. arXiv:1806.03962. https://arxiv.org/abs/1806.03962
-
Ehteshami Bejnordi et al. (2017). Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer. JAMA, 318(22), 2199–2210. doi:jama.2017.14585
