Skip to content
master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
img
 
 
 
 
 
 
 
 
 
 

header

Description

"Histopathologic Cancer Detection" is developed by Team Capybara (gsarti, stinco, andrealorenzon) as final project for the Statistical Machine Learning course held by Prof. Luca Bortolussi at University of Trieste.

The project is based on the Kaggle competition "Histopathologic Cancer Detection", in which participants create an algorithm to identify metastatic cancer in small image patches taken from larger digital pathology scans.

Data

The data are a slightly modified version of the PatchCamelyon (PCam) benchmark dataset in which duplicates generated by probabilistic sampling were removed.

PCam packs the clinically-relevant task of metastasis detection into a straight-forward binary image classification task, akin to CIFAR-10 and MNIST. Models can easily be trained on a single GPU in a couple hours, and achieve competitive scores in the Camelyon16 tasks of tumor detection and whole-slide image diagnosis. Furthermore, the balance between task-difficulty and tractability makes it a prime suspect for fundamental machine learning research on topics as active learning, model uncertainty, and explainability.

The data are provided under the CC0 License. They can be found in the data folder.

Approaches

We compared three approaches for this classification task:

  • Unsupervised segmentation of cellules nuclei followed by a random forest on full-image cell statistics.

  • DenseNet-169 convolutional neural network with pretrained weights and adaptive learning rate, based on the top Kaggle kernel for the challenge.

  • Capsule networks for tumor detection

More information on the project and resources can be found in our project presentation.