Mri segmentation and the quantification of white matter lesions

Magnetic Resonance Imaging (MRI) has rapidly become one of the most important diagnostic tools available to clinicians. This popularity is mainly due to the facts that MRI has no known side-effects, that it is non-invasive, and that it has a very high resolving power for soft tissues. As such, it is extremely valuable for in-vivo studies of the human brain. An additional advantage of MRI is its multispectral character, which makes it possible to generate images of the same physical space with different spectral signatures. The interpretation of MRI data is currently done in a predominantly visual and qualitative fashion. However, the increasing amount of data generated by modern MR scanners and the clinical demand for accurate, reproducible, and quantitative data analysis promote the development of computer-aided techniques. An artificial neural network based image analysis system has been developed and tested, focused on the quantification of so-called white matter lesions in the human brain. Early in the course of this research it became clear that spatial intensity variations, often present in MRI data, hamper the use of classical pattern recognition techniques. Several methods for the correction of this artifact and studies showing the reliability of the proposed segmentation technique are presented.