A deep learning project comparing three CNN architectures for multi-class cancer classification from histopathological images.
This project implements and compares three different Convolutional Neural Network approaches for classifying cancer types from medical images:
| Model | Description |
|---|---|
| Baseline CNN | Custom-built CNN architecture from scratch |
| Improved CNN | Enhanced architecture with optimized hyperparameters |
| Transfer Learning | Pre-trained model fine-tuned for cancer classification |
- Size: 25,000 histopathological images
- Classes: 5 different cancer types
- Source: Academic Torrents
⚠️ The dataset is too large to host on GitHub. Please download it from the link above.
├── Assignment_DL_[Ivaylo Papazov].ipynb # Main notebook with all implementations
├── Deep Learning Assignment Template.pdf # Detailed report and analysis
└── README.md
pip install tensorflow numpy pandas matplotlib scikit-learn- Download the dataset from the link above
- Open the Jupyter notebook
- Update the data path to your local dataset location
- Run all cells to train and evaluate the models
The notebook includes:
- Training/validation accuracy curves
- Confusion matrices
- Comparative analysis across all three architectures
- Hyperparameter tuning experiments
For detailed results and analysis, refer to the PDF report.
- Comparison of model performance across different architectures
- Impact of hyperparameter tuning on classification accuracy
- Effectiveness of transfer learning for medical image classification
- Dataset: LC25000 Lung and Colon Histopathological Image Dataset
Ivaylo Papazov
This project is available for educational purposes.