Bradford, Ian; (2003) Content based image retrieval on radiographs of bone dysplasia syndrome. Doctoral thesis (Ph.D), UCL (University College London).

Text Content_based_image_retrieval_.pdf Download (24MB)

Abstract

Visual clues are an important factor in the diagnosis of many dysmorphic conditions and computer-based resources are becoming increasingly important as diagnostic tools. Although images are a fundamental resource in many such systems, the way in which they are used does not correspond directly to the way in which physicians use images when making a diagnosis. In dysmorphology related systems that use images, images are mapped to textual classifications or descriptions and a semantic gap exists between the information an image conveys and attempts to describe or classify it verbally. Content-Based Image Retrieval (CBIR) is an area of computer research devoted to confronting and solving such issues and the central research topic of this project is the development of CBIR within dysmorphology using real data. To this end a CBIR system was developed. It contained two sub-systems each based on a different approach to CBIR: domain-dependent and domain-independent. The former used a priori medical information and geometric approximation algorithms and the latter used multiple-scale-space-analysis algorithms. A detailed examination of radiographic images from a major dysmorphology resource was undertaken and a sub-set of radiographs selected as a basis for the proof-of-concept system. A method of semi-automatic image segmentation using splines was developed. This was an integral part of the retrieval algorithms and led to the development of specialised systems, including an icon representation system, to provide spline-image functionality. A formal evaluation was conducted to assess the accuracy of the CBIR systems and evaluate User interaction with the Icon System. This used three external data sets as search criteria for the CBIR and online evaluation for the Icon System. Encouraging results were obtained for this successful prototype CBIR system developed around images from the domain of dysmorphology.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item