This is the Github project for machine learning applications on breast cancer diagnosis, created by team 10 of ECE 228, 2022 spring.

Group members:
Haodong Qin: [email protected]
Mingchen Yao: [email protected]
Min Zhong: [email protected]

There are four folders within this project: Diagnostic_data, Self_constructed_CNN, Pre_trained_model and SHAP, which corresponds to the four parts of the project.

Diagnostic_data -- ML methods applied on the diagnostic (tabular) dataset.
Self_constructed_CNN -- VGG-16-like CNN applied on the image dataset.
Pre_trained_model -- Several pretrained CNN models applied on the image dataset.
SHAP -- SHAP analysis on both diagnostic data NN and image data NN.

There are mainly jupyter notebooks for all the three methods, should be easy to run. Make sure you have Pytorch and Scikit-learn installed and the corresponding dataset downloaded.

Also, it would be better to use GPU for the CNN training with the image dataset.

The dataset needed for the analysis is already included in the same directory as DiagnosticData.ipynb. It is part of the Wisconsin Breast Cancer (Diagnostic) Dataset: https://archive.ics.uci.edu/ml/datasets/breast+cancer+wisconsin+(diagnostic)

This is a self-constructed CNN. Please make sure that this directory contains the image data. image data can be downloaded from https://www.kaggle.com/code/mohitgora/cancer-detection-cnn/data. Code is run in jupyter notebook.

Please make sure you have Shap installed. The SHAP analysis itself may take several minutes to complete. Besides, before each analysis, the network need to be trained, which takes ~minutes for diagnostic data and ~ 1 hour for image data. It would be better to use GPU for the CNN training with the image dataset.