DETERMINING SIMILARITY IN HISTOLOGICAL IMAGES USING GRAPH-THEORETIC DESCRIPTION AND MATCHING METHODS FOR CONTENT-BASED IMAGE RETRIEVAL IN MEDICAL DIAGNOSTICS

Abstract

Computer-based analysis of histological images has been gaining increasing attention, due to their extensive use in disease diagnosis and understanding the composition of tissues. This work aims to contribute towards the de-scription and retrieval of histological images by employing a novel structural method based on graphs. Graphs have conquered an influential role for rep-resenting different complex phenomena. Due to their expressive ability, they are considered as a powerful and versatile representation formalism, thereby obtaining a growing consideration especially by the image processing and computer vision community. The thesis describes an innovative method for determining similarity between histological images through graph-theoretic description and match-ing, for the purpose of content-based retrieval. A higher order (region-based) graph representation of breast biopsy images has been attained and a tree-search based inexact graph matching technique has been employed, that fa-cilitates the automatic retrieval of images from a database, structurally similar to a given image. The experimental results obtained and evaluation performed demon-strate the effectiveness and superiority over a simpler histogram-based tech-nique. The method maybe useful for Content Based Image Retrieval (CBIR)-requiring applications in the areas of medical diagnostics and research, and can potentially be generalized for retrieval of different types of complex im-ages.