Meeting Banner
Abstract #1612

Automatic Quantification of 3D Morphology and Appearance of Intervertebral Discs in High Resolution MRI

Ale Neubert1, 2, Jurgen Fripp1, Craig Engstrom3, Duncan Walker4, Raphael Schwarz5, Stuart Crozier6

1The Australian E-Health Research Centre, CSIRO ICT Centre, Brisbane, QLD, Australia; 2School of ITEE, University of Queensland, Brisbane, QLD, Australia; 3School of HMS, University of Queensland, Brisbane, QLD, Australia; 4Wesley Hospital, Brisbane, QLD, Australia; 5Siemens Healthcare, Erlangen, Germany; 6School of ITEE, The University of Queensland, Brisbane, QLD, Australia

High resolution MRI of the spine provides new possibilities for automatic quantification of intervertebral discs (IVDs) morphology and appearance. Discriminative features (planar measurements, features extracted from 3D statistical shape models of lumbar IVDs and from IVDs histograms) were evaluated on a dataset containing IVDs showing early and advanced stages of degeneration. 3D shape features showed some advantages over planar morphological features for detection of early degenerative changes while IVDs in advanced stages of degeneration were best described by features based on signal intensity.

Abstract PDF Poster PDF

Keywords

abnormal achieved advanced advances advantageous although anterior apparent appearance around asymptomatic attempt attenuation automatic automatically back benefit best bottom brighter classification clinical combined commonly component consisted context cross cubic database dataset datasets degeneration describe described describing descriptors detect detection developed disc discs distinct distinctive distributions early evaluate evaluated evaluation example expected extract feature features fold funding ground health healthy height histogram histograms hospital human hypo identified independently inhomogeneity intensities intensity inter isotropic landmarks largely lumbar machine manually marker material maximization mesh metrics middle mixture mode model modes morphological morphology narrowing novel objects overall overlapping pain patients peak peaks performance planar posterior practice previously principal procedure prof projects quantification quantify radiological radiology reflected relevant resolution scheme school sections segmentation segmentations segmented sens separate sets severe shape signs slice software space spec specificity splitting statistical structural studied subjects subset support symptomatic table together tools traditional training truth typically validation variation volumetric width younger