Content-based image retrieval from whole slide histolgy scans

Abstract

This paper presents results that were obtained during solving problem of searching stained regions of tissue. The object of research is whole slide histology scans of tissue sections, that were damaged by cancer. The main idea is transformation of colored image to a gray-scale one, then splitting it to tiles for which different descriptors (intensity histogram, co-occurence matrix) are calculated. The similarity map is obtained by applying comparing of precalculated descriptors of tiles with descriptor of tile that represents damaged region of tissue using metrics (L1, L2, Chi-metric) and visualization using existing color maps. The conclusion was that this methodology can be effectively used to perform query searching fast even on non-specialized computers.