QCIF Home » Industry » Segmentation of Thoracolumbar Muscles

Print version

The thoracolumbar muscles control the dynamic and postural stability of the lower spine. Detailed knowledge on the morphometry of the thoracolumbar muscles is crucial for biomechanical and clinical investigations into low back pain, which affects over 70% of people during their life [1], and injury of the lumbar spine. Increasingly, magnetic resonance imaging (MRI) is being used for morphometric analyses of the thoracolumbar muscles [2-4] although current investigations involve time- and expertise-intensive manual segmentation of the numerous, architecturally complex trunk muscles which span multiple vertebral levels. To enable large-scale, quantitative studies on the thoracolumbar muscles for use in routine clinical settings, automated software for fast and robust segmentation is required for the extraction of accurate and objective morphometric data such as muscle volume and physiological cross-sectional area. This project will develop automated software, implementing novel algorithms based on domain-specific knowledge, for performing objective morphometric analyses and 3D visualisation of the thoracolumbar muscles derived from MRI data. In particular, the algorithms / software will enable rapid automated segmentation of the thoracolumbar muscles, co-registration of serial MR examinations and 3D rendering / visualisation of segmented muscles.  

This project is a collaborative initiative between the Electromagnetics and Imaging (EMI) Research Division (School of ITEE at the University of Queensland) and Southernex Imaging Group. Southernex Imaging Group will provide MRI data and consultancy time of a radiologist to support this project. EMI will provide specialist expertise in medical image analysis, medical instrumentation, and biomedical engineering, and will manage the project. The outcomes of this project will expand the clinical imaging services offered by the industry partner, Southernex Imaging Group and lead to commercially valuable intellectual property.

The Technology

This project will develop novel algorithms/software for automatically: (i) spatially co-registering axial MR images of the thoracolumbar region fromsequential MR examinations of a patient; (ii) segmenting (delineating) the individual thoracolumbar muscles in 3D using deformable contour/surface models; (iii) visualising (rendering) these muscles in 3D; and (iv) extracting quantitative features (measurements) for subsequent morphometric analyses of the thoracolumbar muscles from axial MR images typically used in biomechanical and clinical studies [2, 5-7]. Figure 1 illustrates a representative MR series of T1-weighted axial images of the thoracolumbar muscles used in a major prospective study to discover the highly significant association between asymmetry of the quadratus lumborum muscle and stress fractures in the lumbar spine of adolescent cricket fast bowlers (further details are published in Engstrom et al. 2007, deVisser et al., 2007).

The proposed project is both technically challenging (architectural complexity of the musculature) and computationally demanding (large data volume / dimensionality and deformability of tissues requiring computationally expensive non-rigid registration). With regards to the technical challenges we intend to incorporate domain knowledge (e.g. anatomical knowledge) to ensure accuracy, robustness, and reproducibility of the results. With respect to the computational demands we propose to use high performance computing and/or grid computing; e.g. a multiprocessor high performance computer, such as QCIF’s 16 processor SGI Altix, can be used to perform non-rigid registration of individual pairs of axial slice images in parallel. The grid-computing paradigm is particularly attractive because it offers the possibility of uploading data acquired in the clinical setting to a computational grid where the computationally demanding registration and segmentation procedures are performed, and a short time later the spatially aligned and segmented data are downloaded for 3D visualisation and morphometric analyses on a PC.

 

Participants

Dr Craig Engstrom,  Dr Andrew Mehnert
School of Human Movement Studies and School of ITEE, University of Queensland

Industry Participants

Dr Duncan Walker
Southernex Imaging Group
Wesley Hospital

Reports

Final Report - May 2008 (864 KB PDF)
Progress Report - November 2007 (86 KB PDF)
Progress Report - August 2007 (288 KB PDF)
Project Proposal  (230KB PDF)