Clinically based computerized analysis of magnetic resonance imaging of the spine

Neubert, Ales (2013). Clinically based computerized analysis of magnetic resonance imaging of the spine PhD Thesis, School of Information Technology and Electrical Enginering, The University of Queensland.

Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
s4240557_phd_thesis.pdf Thesis (fulltext) application/pdf 13.69MB 9
Author Neubert, Ales
Thesis Title Clinically based computerized analysis of magnetic resonance imaging of the spine
School, Centre or Institute School of Information Technology and Electrical Enginering
Institution The University of Queensland
Publication date 2013
Thesis type PhD Thesis
Supervisor Stuart Crozier
Jurgen Fripp
Total pages 170
Language eng
Subjects 0803 Computer Software
0903 Biomedical Engineering
0801 Artificial Intelligence and Image Processing
Formatted abstract
Spine-related disorders account for the largest proportion of musculoskeletal complaints in industrialized countries. Magnetic Resonance (MR) imaging provides an excellent diagnostic tool for assessing common spine pathologies, including, disc prolapse or herniation, vertebrae or disc degeneration, and osteoporosis. Traditionally, the radiological spine exam consists of acquisition of several multi-slice two-dimensional (2D) MR scans with thick slices, usually in the sagittal and axial directions. Recent advances in MR hardware, software and MR sequence design introduce new possibilities in musculoskeletal radiology, such as acquisition of high resolution three-dimensional (3D) scans. Acquisition of contiguous thin slices has been shown to improve the depiction and visualization of anatomical structures, to increase diagnostic sensitivity and the 3D images can be reformatted to visualize the acquired volume in any plane, therefore reducing the acquisition time. Although these images visualize the spine tissues in higher resolution, they also increase the amount and complexity of the data that has to be processed in clinical settings or research trials, requiring devotion of significant radiological resources and specialist time.

Efficient informatics tools offer considerable opportunities for radiological assessment of high resolution volumetric data. Computer-Aided Diagnosis (CAD) devices have been used in many clinical procedures to alleviate time-consuming and tedious tasks, such as the segmentation of anatomical features, morphological assessment or region-of-interest delineation. Previous spine CAD systems have predominantly focused on 2D MR, computed tomography (CT) or X-ray imaging modalities, while the potential of automated analysis of the relatively novel high resolution spine MR data remain largely unexplored. One accountable factor is the fact that the volumetric nature of the novel MR data challenges existing automated spine processing techniques due to the high resolution and the underlying imaging artifacts.

The goal of this research is to develop sophisticated image processing techniques for the new 3D MR acquisition protocol. These tools will deliver accurate and precise quantitative data that can help the clinical assessment of common spine disorders. To this end, the specific aims of this research were: (1) develop an algorithm for automated 3D segmentation of vertebral bodies (VBs) and intervertebral discs (IVDs) from high resolution 3D MR scans; (2) extract accurate and reproducible quantitative morphological measurements form the MR scans; and (3) use the segmentation algorithm and the extracted measures to design a CAD system for the detection of a selected spine pathology. In this thesis, we have focused on the clinical example of the degenerative disc disease due to its high prevalence and unclear relationships to patient’s clinical symptoms, and therefore potentially benefiting from reproducible and objective computer-aided assessment.

An innovative automated segmentation algorithm combining object recognition techniques for spine localization and active shape models for deformable segmentation was developed (Aim 1). A subset of 14 cases was manually segmented and used for quantitative assessment. The method was applied to MR images of 28 subjects acquired with the T2-weighted 3D SPACE sequence and competitive results were obtained for both IVDs and VBs. The presented technique outperforms existing 3D MR segmentation algorithms for VBs in accuracy, while it is the first time that an intrinsically 3D approach is presented for segmentation of IVDs from volumetric scans.

A study on reliability and validity of automatically extracted IVD measurements (height and volume) was performed using manual measurements on MR images of 7 subjects acquired under three conditions (baseline, pre-exercise, post-exercise) (Aim 2). Strong test re-test reproducibility (baseline, pre-exercise) was found for both methods (computerized and manual), and good concurrent agreement was found. Both methods detected significant longitudinal changes following 30 minutes of moderate intensity exercise. The computerized method therefore provides accurate volumetric IVD measurements for analysis of large cross-sectional, interventional or longitudinal studies, not feasible with traditionally used manual techniques.

A CAD system for classification of degenerated IVDs from T2-weighted 3D SPACE and routine clinical 2D turbo spin echo MR images was developed (Aim 3). The classification is based on novel 3D morphological and signal intensity features and the approach was evaluated on two datasets: asymptomatic volunteers (28 subjects) and symptomatic patients (11 cases). The approach was successful in detecting both early and advanced IVD degeneration in both image types and showed that the proposed 3D morphological features offer significant advantage over previously used 2D morphological features in CAD of the degenerative disc disease.

The results show that the high resolution volumetric MR scans contain salient anatomical and diagnostic information, and reveal the potential of the proposed innovative computerized tools for automated processing of spine MR images. The technical developments described in this thesis can provide new insights for large-scale pathogenic studies of spine disorders and supply an objective method that can be seamlessly integrated into the clinical workflow. 
Keyword Spine
Intervertebral Disc
Vertebral body
Magnetic Resonance Imaging
Statistical shape modeling
Computer aided diagnosis
Degenerative Disc Disease
Feature extraction

Citation counts: Google Scholar Search Google Scholar
Created: Wed, 30 Apr 2014, 10:10:59 EST by Mr Ales Neubert on behalf of Scholarly Communication and Digitisation Service