Automating breast cancer diagnosis in histopathology

• Dataset was provided by ICIAR2018 Challenge.
• Data comprised by 400 images [1535x2048x3] of 4 classes (100 each): Benign, InSitu, Invasive, Normal.

Two approaches developed:

1. Traditional machine learning: Extract Fisher Vector representation from histopathology images and train a SVM classifier.
2. Deep learning: Learn representation automatically using a Convolutional Neural Network (CNN) in Pytorch (to be updated).

Jupyter notebooks:

1) Reading and visualising images

1-Data_Visualisation.ipynb

2) Classification: FisherVector+SVM (80% test acc.)

2-FisherVector_SVM.ipynb