In this project a novel model based scheme of image interretation was developped.
For further information contact:
|
|
|
In this project a novel model based scheme of image interretation was developped. For further information contact: |
|
|||||
| David Pycock | Tel.: | +(0)121 414 4330 | Email: | D.Pycock@bham.ac.uk | ||
| Background | |||
| The volume and shape of a 3-D structure is important in medical diagnosis. Whilst some structures, such as the major organs of the body and the ventricles of the brain, have a relatively well defined boundary in CT X-Ray and MR images there are a many cases where the shape is complex and the boundary is not clearly defined. The scar left by a stroke and the hippocampus are examples of such structures in the brain. Ultrasound images are also characterised by a strong speckled pattern which are difficult to visualise unaided. The technique for detecting boundaries described here would be valuable in many such cases, in addition to the analysis of stroke lesions. | |||
| Objectives | |||
 |
To define quantitative measures for use in clinical protocols for assessing stroke patients. |
|
To simplify measurement of the volume of stroke lesions. |
|
To develop a robust process for delineating complex 3-D boundaries. |
| Previous Research | ||
| Existing research
involves the use of active contours to compute an "optimal" boundary in
each of a series of slices and ways of computing 3-D boundaries using deformable
models with energy minimisation procedures. Each of these methods produces
a good interpretation of convex shapes but none can accurately identify
the boundary of highly convoluted shapes. Fritisch et al. are adapting
the 2-D cores algorithm developed at the University of North Carolina to
compute 3-D boundaries by modelling 2-D structures on a "tiltable" plane.
In the School of Electronic and Electrical Engineering we have developed a novel technique for 2-D medical image interpretation which is robust and able to identify low contrast boundaries and improved procedures for computing grey-level symmetry. This project involves extending and integrating these techniques for the analysis of 3-D structures. |
||
| Proposal | |||
| The intention
in this project is to use a multi-resolution scale-space representation
to minimise noise sensitivity and maximum likelihood criteria to maximise
the sensitivity of the boundary detection process. By using axes of symmetry
from 3-D data sets the intention is to avoid dependence on 2-D constraints
The symmetries of 3-D volumes will enable surface orientation and object size to be estimated before the boundary has been detected. This information will be used to guide the boundary detection process in which weak models of correlation between neighbouring surface points will be combined with strong models that define surface discontinuity. Evaluation will be based, in part, on an existing database of images. |
|||
| Facilities and Benefits | |||||
| Extensive computing resources and a well stocked library are available to support this research which is expected to lead to publications in international journals. There will be an opportunity to submit results for presentation at national and international conferences. The analytical and computing skills that you will develop are much sought after by potential employers. | |||||