Modelling and Matching 3D Objects for Medical Image Analysis
Lead Research Organisation:
University of Manchester
Department Name: Medical and Human Sciences
Abstract
Over recent years there has been a huge increase in the number of volumetricimages generated in the medical domain. PET, MR and CT machines are becomingmore common, and are essential tools to investigate the anatomy and functionof the body. However, interpretting 3D images is difficult. Althoughradiologists are very good at making qualitative judgements, to makeaccurate quantitative measurements requires careful annotation ofthe structures of interest in a 3D volume, which is hard to achieve.There has thus been an extensive research effort to develop software toolsto help extract useful information from such data. These range fromsimple image annotation tools (to aid manual segmentation) to fullyautomatic systems which (ideally) require no manual intervention.Although significant progress has been made, there are still manydifficult problems which must be tackled before automatic systems canproduce really reliable results and are suitable for wide application.This project seeks to address such issues.Our group has pioneered a number of methods of constructing statistical models of the shape and appearance of structures within the body. These describe the way such structures vary in shape across healthy and diseased subjects, and can be rapidly matched to new images, allowing one to determine the shape of the structure in that image. These methods have been adopted by both researchers and companies world-wide, and have been very influential. However, it currently takes significant time and skill to apply them to new problems. This project aims to develop methods which will make it much simpler to use the models in new areas, and will produce more accurate and robust results.The new methods will be evaluated by using them on three different structures - knees, kidneys and organs within the brain. It is anticipated that the methods will allow more widespread adoption of the current model-based techniques, and will have an impact in both clinical and pharmaceutical research.
Organisations
Publications
Babalola KO
(2009)
An evaluation of four automatic methods of segmenting the subcortical structures in the brain.
in NeuroImage
Babalola KO
(2008)
3D brain segmentation using active appearance models and local regressors.
in Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
Cootes TF
(2010)
Computing accurate correspondences across groups of images.
in IEEE transactions on pattern analysis and machine intelligence
Talbot PS
(2010)
Brain serotonin transporter occupancy by oral sibutramine dosed to steady state: a PET study using (11)C-DASB in healthy humans.
in Neuropsychopharmacology : official publication of the American College of Neuropsychopharmacology
| Description | Image segmentation - delineating structure(s) of interest in an image, is a fundamental part of medical image processing. For instance, it is often important to find the surfaces of bones or organs, either to analyse their shape or to make measurements of the patterns of image intensity withing the object. As imaging becomes more routinely used for diagnostic and therapeutic purposes, automating segmentation is necessary to increase throughput of processing, reduce the workload on staff involved and increase the quality of quantitative information derived from segmentation. This project addressed the automation of both the construction of 3D volumetric Active Appearance Models (AAMs) and their use to segment structures in medical images. AAMs include statistical models of the shape and appearance of objects or collections of objects, and use an efficient algorithm to match such models to a new image, allowing accurate segmentation of the objects. During the project we focussed on: a) Improved methods of finding correspondences across groups of images - a critical component of the model building process. We developed 3D "groupwise" registration techniques, which automatically find a way to deform each image in a set so as to best match to a mean reference image. b) Automatic construction of models which decompose complex objects into separate parts, together with geometric models of their relative position. These models are effective for locating the approximate position of structures in an image, and so serve to initialise more complex models. We showed that this approach leads to better results than previous, simpler, methods of initialisation. c) Improving the accuracy of segmentation by novel methods of voxel classification, using the results of Active Appearance Model matching to initialise local models which decide which voxels belong to which structure of interest. |
| Exploitation Route | The groupwise registration methods developed during the project are being used by a spin-off company, Imorphics Ltd (Manchester), to build statistical models of appearance which they use in medical image analysis. They particularly focus on a range of musculo-skeletal problems, working with pharmaceutical companies to evaluate the efficacy of drugs on various forms of arthritis, and on understanding the structure of various joints and the way in which they degrade during disease. The groupwise registration methods developed during the project are being used by a spin-off company, Imorphics Ltd (Manchester), to build statistical models of appearance which they use in medical image analysis. They particularly focus on a range of musculo-skeletal problems, working with pharmaceutical companies to evaluate the efficacy of drugs on various forms of arthritis, and on understanding the structure of various joints and the way in which they degrade during disease. One output of the project is a tool to use the 3D Active Appearance Models to automatically segment 10 subcortical structures in 3D MR images of the brain. This has been made publicly available (http://www.isbe.man.ac.uk/~kob/vaam_1_0/index.html). This is being used in a variety of research projects. |
| Sectors | Healthcare,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology |
| URL | http://personalpages.manchester.ac.uk/staff/timothy.f.cootes/Projects/Model3D/model_building_and_matching.html |
| Description | Ideas and techniques developed in the project have been used to help develop an automatic system for analysing CT images of chickens, in collaboration with a large breeding company. A system has been licenced to the company. |
| First Year Of Impact | 2017 |
| Sector | Agriculture, Food and Drink |