/camelyon

Classification of WSIs from the Camelyon dataset in Pytorch

Primary LanguageJupyter Notebook

Whole Slide Histopathology Classification

Automated detection and classification of WSIs from the Camelyon dataset in Pytorch, used for class project.

Batch Generator

WSIs are large images, so standard deep learning classification methods won't work, because it isn't possible to feed a 10 billion pixel image into a CNN. With WSIs the image is broken up into smaller tiles, and a classifier is used to classify these. This classifier can be applied across the entire image, generating a heatmap of cancer probability. These tiles can either be generated within the pytorch batch generator, or created and saved:

Tile Level Classification

This focuses on the tile-level classification, with some investigation into ensemble methods for classification. Bagging was shown to be ineffective, and other ensembles were somewhat more effective.