Shape-based myocardial contractility analysis using multivariate outlier detection

  • Authors:

  • Karim Lekadir;Niall Keenan;Dudley Pennell;Guang-Zhong Yang

  • Affiliations:

  • Visual Information Processing, Department of Computing, Imperial College London, UK;Cardiovascular Magnetic Resonance Unit, Royal Brompton Hospital, London, UK;Cardiovascular Magnetic Resonance Unit, Royal Brompton Hospital, London, UK;Visual Information Processing, Department of Computing, Imperial College London, UK

  • Venue:
  • MICCAI'07 Proceedings of the 10th international conference on Medical image computing and computer-assisted intervention
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents a new approach to regional myocardial contractility analysis based on inter-landmark motion (ILM) vectors and multivariate outlier detection. The proposed spatio-temporal representation is used to describe the coupled changes occurring at pairs of regions of the left ventricle, thus enabling the detection of geometrical and dynamic inconsistencies. Multivariate tolerance regions are derived from training samples to describe the variability within the normal population using the ILM vectors. For new left ventricular datasets, outlier detection enables the localization of extreme ILM observations and the corresponding myocardial abnormalities. The framework is validated on a relatively large sample of 50 subjects and the results show promise in localization and visualization of regional left ventricular dysfunctions.