Simmons, Andrew Mark Kester; (1992) Segmentation of neuroanatomy in magnetic resonance images. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Segmentation_of_neuroanatomy_i.pdf Download (22MB) | Preview

Abstract

Segmentation in neurological Magnetic Resonance Imaging (MRI) is necessary for volume measurement, feature extraction and for the three-dimensional display of neuroanatomy. This thesis proposes several automated and semi-automated methods which offer considerable advantages over manual methods because of their lack of subjectivity, their data reduction capabilities, and the time savings they give. Work has concentrated on the use of dual echo multi-slice spin-echo data sets in order to take advantage of the intrinsically multi-parametric nature of MRI. Such data is widely acquired clinically and segmentation therefore does not require additional scans. The literature has been reviewed. Factors affecting image non-uniformity for a modem 1.5 Tesla imager have been investigated. These investigations demonstrate that a robust, fast, automatic three-dimensional non-uniformity correction may be applied to data as a pre-processing step. The merit of using an anisotropic smoothing method for noisy data has been demonstrated. Several approaches to neurological MRI segmentation have been developed. Edge-based processing is used to identify the skin (the major outer contour) and the eyes. Edge-focusing, two threshold based techniques and a fast radial CSF identification approach are proposed to identify the intracranial region contour in each slice of the data set. Once isolated, the intracranial region is further processed to identify CSF, and, depending upon the MRI pulse sequence used, the brain itself may be sub-divided into grey matter and white matter using semiautomatic contrast enhancement and clustering methods. The segmentation of Multiple Sclerosis (MS) plaques has also been considered. The utility of the stack, a data driven multi-resolution approach to segmentation, has been investigated, and several improvements to the method suggested. The factors affecting the intrinsic accuracy of neurological volume measurement in MRI have been studied and their magnitudes determined for spin-echo imaging. Geometric distortion - both object dependent and object independent - has been considered, as well as slice warp, slice profile, slice position and the partial volume effect. Finally, the accuracy of the approaches to segmentation developed in this thesis have been evaluated. Intracranial volume measurements are within 5% of expert observers' measurements, white matter volumes within 10%, and CSF volumes consistently lower than the expert observers' measurements due to the observers' inability to take the partial volume effect into account.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item