Breast-Cancer-MaskRCNN

This is the implementation of my bachelor thesis Breast Cancer Diagnosis Based on Artificial Intelligence with Mammography supervised by Kun Cheng. This is a breast cancer diagnosis system based on Mask RCNN with mammography. In the thesis, I explored different aproaches on mammography segmentation. Given a mammography, the system is supposed to:

locate the masses (object detection)
classify the pathology (malignant/benign) of the masses (image classification)
segment the mass (instance segmentation)

An example is:

The code is based on the pytorch implementation of Mask-RCNN

Dataset

Download from CBIS-DDSM
Run dataset.py. The dataset will be organized as the file structures shown below.

Train

Run mammography/Extract_metadata_to_JSON.ipynb to get data description file(s). Put the file(s) at dataset/mammo/ (your dataset file).
Run mammography/train.ipynb.

Test

Run mammography/evaluate.ipynb.

Finetune

If you use the model and data provided by this work, you will get 0.7 on mass detection sensitivity and 0.53 on mass segmentation IOU. Please tune the parameters and train more epoch to get a better model.

Tips

The plot loss cube in training does not work well, please use the tools in loss/ instead.