TDAExplore: quantitative image analysis through topology-based machine learning

Abstract

Machine learning has greatly expanded the ability to classify images. However, many machine learning classifiers require thousands of images for training and lack quantitative descriptors of how images were grouped. We overcome these limitations with a machine learning approach based on topological data analysis, where a data set of 20-30 images is sufficient to accurately train the classifier. Our method quantifies differences between groups and identifies subcellular regions with the largest dissimilarities.

Topology-based fluorescence image analysis for automated cell identification and segmentation

“…TOpological Boundary Line Estimation using Recurrence Of Neighbouring Emissions (TOBLERONE) enables identification of intensity modes within images by means of the topological data analysis techniques known as persistent homology and mode seeking. Topological image analysis is itself a relatively young field, which has shown promise in probing microbiology [10]. Similar algorithms have used this form of analysis to segment nuclei in histological slides of liver tissue [11].…”