gms | German Medical Science

GMDS 2013: 58. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V. (GMDS)

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie

01. - 05.09.2013, Lübeck

Combined Registration and Segmentation of the Left Ventricle in Cine MR Image Data

Meeting Abstract

Suche in Medline nach

  • Timo Kepp - Universität zu Lübeck, Institut für Medizinische Informatik, Lübeck, DE
  • Jan Ehrhardt - Universität zu Lübeck, Institut für Medizinische Informatik, Lübeck, DE
  • Heinz Handels - Universität zu Lübeck, Institut für Medizinische Informatik, Lübeck, DE

GMDS 2013. 58. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e.V. (GMDS). Lübeck, 01.-05.09.2013. Düsseldorf: German Medical Science GMS Publishing House; 2013. DocAbstr.334

doi: 10.3205/13gmds265, urn:nbn:de:0183-13gmds2651

Veröffentlicht: 27. August 2013

© 2013 Kepp et al.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen (http://creativecommons.org/licenses/by-nc-nd/3.0/deed.de). Er darf vervielfältigt, verbreitet und öffentlich zugänglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.


Gliederung

Text

Introduction: Magnetic resonance imaging is a non-invasive and radiation-free modality, which offers spatial and temporal image data acquisition. To deal with the resulting increased number of data (semi-)automatic image processing methods are required. With our approach, we present a method for combined registration and segmentation of the left ventricle (LV) and a motion field calculation of the heart in spatial and temporal image data. There have been published some former works of combined registration and segmentation [1], [2]. Schmidt-Richberg et al. used a non-linear diffusive registration combined with level set segmentation, which we will also introduce in this work.

Material and Methods: The aim of the proposed approach consists of two subparts: The extraction of myocardial tissue of the LV and the computation of deformation fields for heart motion characterization. For this work, we use a spatial temporal short axis cardiac MR image dataset, wherein time domain enfolds a complete cardiac cycle. Due to large spacings between the image slices, gradient calculations are difficult to determine and smooth segmentation results can't be achieved. Therefore, we interpolate the image data by using a registration-based interpolation scheme [3]. For our approach, we search for segmentations of the myocardium and transformations that represent the motion of the LV for all time steps. Assuming a given initial segmentation of the myocardium at a reference time step, one can use non-linear registration to transfer the given segmentation to all other time steps. Otherwise, level set segmentation can be used to calculate segmentations of all time steps directly. In our combined approach, a similarity condition establishes a link between registration and segmentation. Here, we introduce a shape prior term, which avoids long distances between both segmentation results and prevents a contour leakage during segmentation step.

Results and Discussion: For validation, stand-alone registration and stand-alone segmentation as well as the combined algorithm are compared by their segmentation results. We use MR data that consist of five different patient data sets, provided by Andreopoulos et al. [4]. In addition, the data set enfolds manual generated segmentations, which are used as gold standard. Validation results demonstrate an improvement of segmentation result by the combined algorithm (mean surface distance: msd = 2,48mm; Hausdorff distance: H = 3,09mm). Single registration also provides advanced segmentation results (msd = 2,73mm; H = 7,82mm), but is not as robust as the combined algorithm. In contrast, single segmentation provides worse results (msd = 6,74mm; H = 21,3mm). One reason are low gray value differences in target tissue, which leads to a rapid contour leakage to surrounding areas.

Conclusion: We introduced an approach for combined non-linear registration and level set segmentation in this abstract. In reference to medical context, we demonstrated an improvement for semi-automatic segmentation and motion field estimation of the LV. Combined approach shows good results in spite of weak boundary conditions of target tissue in MR image data. Furthermore, we demonstrated an improvement of segmentation result in contrast to single registration and segmentation.


References

1.
Paragios N, Rousson M, Ramesh V. Knowledge-based Registration & Segmentation of the Left Ventricle: A Level Set Approach. Applications of Computer Vision. 2002:37-42.
2.
Schmidt-Richberg A, Handels H, Ehrhardt J. Integrated Segmentation and Non-linear Registration for Organ Segmentation and Motion Field Estimation in 4D CT Data. Methods of Information in Medicine. 2009;48:344-349.
3.
Ehrhardt J, Säring D, Handels H. Structure-preserving Interpolation of Temporal and Spatial Image Sequences Using an Optical Flow-based Method. Methods of Information in Medicine. 2007;46:300-307.
4.
Andreopoulos A, Tsotsos JK. Ecient and Generalizable Statistical Models of Shape and Appearance for Analysis of Cardiac MRI. Medical Image Analysis. 2008;12: 335-357.