This repository contains a comparative study between Convolutional Neural Networks (CNN) and Scattering Networks (ScatNet) for histopathological image classification of lung cancer tissues.
The project aims to classify lung cancer histopathological images using two different deep learning approaches:
- CNN: A custom architecture with minimal parameters (~14K) achieving 99.26% ± 0.72% accuracy
- ScatNet: A wavelet-based approach with fixed mathematical representations achieving 92.99% ± 1.59% accuracy
Key features:
- Binary classification between adenocarcinoma and benign tissue samples
- K-fold cross-validation with 10 folds
- Comprehensive model interpretability analysis
- Advanced visualization techniques for model decisions
- Python 3.11
- 8GB VRAM
- Install dependencies:
pip install -r requirements.txt-
Download the dataset from Kaggle and place the zipped file in the root directory.
-
For interactive exploration, use Jupyter notebook in
main.ipynb:
Report and presentation are available in the doc folder.