1.1 Motivation

next up previous contents
Next: 1.2 Goals Up: 1 Introduction Previous: 1 Introduction

1.1 Motivation


Magnetic resonance imaging (MRI) is a noninvasive method for producing three-dimensional (3D) tomographic images of the human body. MRI is most often used for the detection of tumors, lesions, and other abnormalities in soft tissues, such as the brain. Clinically, radiologists qualitatively analyze films produced by MRI scanners.

Recently, computer-aided techniques for analyzing and visualizing magnetic resonance (MR) images have been investigated. Many researchers have focused on detecting and quantifying abnormalities in the brain. Automatically identifying the brain in MR images of the head is an important step in this process. Another important step for computer-aided analysis is data quality assurance. MR images contain unwanted intensity variations due to imperfections in MRI scanners. Removing or reducing these variations can improve the accuracy of automated analysis.

This thesis presents a novel, fully automatic method for intracranial boundary detection and intensity correction in MR images of the head. The intracranial boundary is the boundary between the brain and the intracranial cavity. It accurately segments the brain from other features in the head.

1.1.1 Multiple Sclerosis Lesion Segmentation


Multiple sclerosis (MS) is an autoimmune disease characterized by damage of the myelin covering of neurons in cerebral white matter. The damaged areas or lesions are distinctly visible in MR images. For this reason, MRI is used to detect and track MS lesions in the brain. As MS progresses, the number and size of lesions in the sufferer's brain increases. During treatment and clinical studies, doctors use MRI scans to monitor this change in lesion volume. The monitoring process involves painstakingly outlining every lesion in scans of possibly hundreds of patients.

Many researchers are investigating methods for automatically segmenting MS lesions in MRI scans of the head [35][25][55][22]. We are currently investigating the method developed by Johnston et al [22].

Johnston's segmentation method uses a stochastic, Bayesian relaxation based algorithm called iterative conditional modes (ICM). Using tissue intensity histograms from manually identified regions in the MR scan, it attempts to isolate all tissues in the brain, including MS lesions, white matter, grey matter, and cerebral spinal fluid (CSF). Theoretically, it can also isolate tissues outside the brain, such as the eyes, skin, fat, and the skull.

Because the distinct tissue types are represented by non-unique intensities in the MR scans, the ICM algorithm confuses tissues outside the brain with tissues inside the brain. For example, the intensity of the eyes could be identical to the intensity of some MS lesions. Further, the speed of the algorithm is proportional to the number of tissues it segments. It is therefore desirable to isolate the brain in MR images before applying the ICM algorithm.

The ICM algorithm assumes that a given tissue type is represented by similar intensities everywhere in the MRI volume. Intensity variations inhibit the algorithm considerably. Consequently, intensity variation due to scanner imperfections must be reduced or removed from the MRI volumes. For reasons discussed in Chapter 2, this reduction of unwanted intensity variation is referred to as radio frequency (RF) correction.

1.1.2 Automatic Registration

In the context of medical imaging, registration refers to the process of spatially aligning two or more scans of the same patient. For MS treatment studies, it is useful to register MRI scans of a patient's head taken at different times in order to track the activity of lesions. The scans, however, need not be of the same modality. For example, it might be useful to register a positron emission tomography (PET) scan, which shows brain activity, with an MRI scan, which shows anatomy.

Absolute points of reference are necessary for successful registration in surface based methods [38]. In order for two independently acquired scans to be accurately aligned, identical features must first be identified in both [15][56][38]. In the absence of surgically implanted markers, the intracranial boundary provides the best feature of reference [15]. Therefore, automatic intracranial boundary detection is an essential step in such registrations.

1.1.3 Image Compression

A single clinical MRI scan occupies several megabytes of disk space (see Chapter 8). Effective image compression schemes are important for storing multitudes of scans. With the advent of teleradiology, where MRI scans are transmitted by wire to remote sites for evaluation by specialists, MR image compression plays a huge role in increasing transmission speed [43][13].

In MRI scans of the head, doctors are usually more interested in the brain as opposed to the region outside the brain. For this reason, Anderson has developed a lossy MRI compression scheme that selectively compresses the region outside the brain at a higher compression ratio than the brain [2]. Thus, he achieves high compression ratios while maintaining image quality in the brain area. Obviously, automatic intracranial boundary detection is a prerequisite for such a scheme.

next up previous contents
Next: 1.2 Goals Up: 1 Introduction Previous: 1 Introduction

Blair Mackiewich
Sat Aug 19 16:59:04 PDT 1995