Robust Systems for Automated Analysis of Structures in 2D Medical Images

点击翻译按钮获取中文摘要

This project aims to develop robust and accurate systems for locating the outlines of bones and other structures in widely used medical images such as radiographs. A key motivation is to provide a set of tools for the clinical research community to help analyse bone shape and thus better understand, monitor and treat musculoskeletal disease. Such diseases affect about 16% of all adults and more than 30% in the over 65s. The annual total cost of arthritis alone is estimated to be more than £30 billion for the UK. There is increasing clinical interest in studying bone shape, but progress is hampered by the time required to annotate the outlines of structures of interest on the large databases now available.We have recently developed new machine-learning based methods for fitting statistical shape models to images which are achieving state-of-the-art performance. In this project we will build on this approach. Our goal is to provide the clinical research community with a system to allow analysis of large medical image databases as efficiently as possible. To achieve this we will address fundamental research challenges including (i) How to ensure model matching is robust and (ii) How to construct models from un-labelled images efficiently, with minimal human interaction. The project will produce both a practical and useful system for clinicians, and new algorithms and insights into tackling fundamental problems in computer vision and medical image analysis.