A QuadTree Image Representation for Computational Pathology

TL;DR

This paper introduces a novel quadtree-based image representation for computational pathology, enabling highly accurate classification with significantly less data compared to existing methods.

Contribution

It is the first to apply quadtrees to pathology images, providing an interpretable and data-efficient approach for downstream classification tasks.

Findings

Achieves comparable accuracy to tissue mask patch methods

Uses over 38% less data

Provides an interpretable image representation

Abstract

The field of computational pathology presents many challenges for computer vision algorithms due to the sheer size of pathology images. Histopathology images are large and need to be split up into image tiles or patches so modern convolutional neural networks (CNNs) can process them. In this work, we present a method to generate an interpretable image representation of computational pathology images using quadtrees and a pipeline to use these representations for highly accurate downstream classification. To the best of our knowledge, this is the first attempt to use quadtrees for pathology image data. We show it is highly accurate, able to achieve as good results as the currently widely adopted tissue mask patch extraction methods all while using over 38% less data.