Computer-Based Quantification of Abnormalities in 3D Brain Images

Introduction
Medical imaging has become one of the primary methods available to the medical profession for the investigation and diagnosis of disease. Within Medical images, it is possible to detect subtle changes resulting from the disease, which may otherwise go undetected. However, because some changes can be very small, even with imaging it can be difficult for the human eye to detect such subtleties. This is where the area of machine vision can aid the medical world.

Dementia is an example of a disease that can most easily be
diagnosed by viewing the images resulting from a Magnetic Resonance scan of the patient. The dementia project has the long-term aim of creating a computer vision based system that can detect the subtle losses (or atrophy) within the brain which occur with the onset of one of the three main types of dementia – Alzheimer’s Disease, Fronto-Temporal Dementia and Vascular Dementia. These changes are to be detected by the quantification of certain regions of the brain, which vary depending on the type of dementia. This quantitative analysis should provide the necessary information upon which a diagnosis can be based.

Current Work
The first part of the project, and our current short-term aim, is to produce some software that can categorize the image into three areas – the brain, the cerebrospinal fluid (CSF) and the skull (including bone, skin, muscle, etc.).

This will then enable the segmentation of the brain into its
constituent parts, enabling the system to look for atrophy. To
achieve this initial categorization it is necessary to be able to search through the image, and identify the boundaries between the skull and the CSF, and the CSF and the brain. This is done using the active shape models developed within this department.

Active Shape Models
Active shape models are statistical models of the shape and gray levels of an image based on points distributed around the image. The models are allowed to deform within certain constraints, and so attempt to find a shape within the image that matches the parameters defined. The parameters are defined by ‘training’ the model on a training set of images, thereby showing the model the shape it should try to detect.

We are currently using such models to highlight the inside of the skull, and the brain stem, as shown in the image.

The Next Step
The next step in the project is to extend the current active shape model into 3D so that the whole of the head can be categorized as explained above. This will enable the quantification of each of the three main categories, and will also enable the extraction of the brain for further segmentation for the quantitative analysis. The project should then be at a stage where the results of the quantification can be used to aid the diagnostic process.

Summary
The overall aim of the project is to use computer vision techniques to aid in the diagnosis of dementia patients. This is to be achieved using quantitative analysis on the Magnetic Resonance images already used for such diagnosis.

The computer vision techniques used are based on the work already done in this department on shape searching within images. This involves using active shape models to identify structures within the brain, and extending them to work in a 3D environment.

Acknowledgements
The funding for this project is from the EPSRC.

Chris Wolstenholme : chris.wolstenholme@.man.ac.uk

http://www.wiau.man.ac.uk

Research Courses Clinical Radiology Industrial liaison Contact Us Join Us Search Home

Top Page