Automated segmentation of human cervical-supraclavicular adipose tissue in magnetic resonance images

Figures

• 

  Figure 1. Flow charts of the semi-automated reference segmentation (REF) method and the automated multi-atlas segmentation (MAS) method. In the MAS method, the initial crude manual delineation of the REF method was replaced by a crude multi-atlas segmentation. Thereafter, a common set of automated fine adjustments was applied to the two methods (thresholding, erosion). In the MAS method, the segmented depot was obtained after a refinement step where residual voxels of air and bone were removed (by thresholding, dilation, filling of holes, erosion). In the REF method, the segmented depot was obtained directly after the automated fine adjustments as voxels in bone and air were assumed to be avoided in the manual delineation.

• 

  Figure 2. Axial fat fraction (FF) maps of the subjects in the multi-atlas segmentation (MAS) group and the evaluation (E) group. Both groups were composed of adults and children with body compositions ranging from lean to obese, here ordered according to BMI.

• 

  Figure 3. Examples of segmentation results from the semi-automated reference segmentation (REF) method and the automated multi-atlas segmentation (MAS) method. Axial fat fraction (FF) maps of two of the 16 evaluation (E) subjects A and B are overlaid with (a) the crude manual delineation (red contour) and the final manual segmentation after fine adjustment (sBATREF, yellow area), (b) the crude multi-atlas segmentations (multi-colour contours) and (c) the fused crude multi-atlas segmentation (blue contour) and the final multi-atlas segmentation after fine adjustment and refinement (sBATMAS, green area). Subject A: female, age = 16 years, BMI = 35.2 kg/m2. Subject B: female, age = 11 years, BMI = 28.6 kg/m2.

• 

  Figure 4. The preprocessing steps applied to the water fraction (WF) map prior to image registration in the multi-atlas segmentation (MAS) method. Two slices of the same subject are illustrated. Step a: Calculation of the WF map from the fat fraction (FF) map. Step b: Removal of background voxels. Step c: Removal of skin voxels. After preprocessing, the WF map consisted of an apparent lean tissue body. As the preprocessed WF map nearly surrounded the cervical-supraclavicular (sBAT) depot and exhibited relatively low inter-subject variability, it was suitable for registration purposes.

• 

  Figure 5. The preprocessing steps applied to the R2 * map prior to image registration in the multi-atlas segmentation (MAS) method. Two slices of the same subject are illustrated. Step a: Removal of background voxels outside of the body and skin voxels. Step b: Reduction of image noise using a median-filter. The preprocessed R2 * map exhibited high voxel intensities in areas close to bone and air-filled structures, which guided the matching of these anatomies under image registration.

• 

  Figure 6. Correspondence in fat fraction, R2 * and volume measurements. Significant correlations observed between (a) mean fat fraction (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{{\rm{FF}}}$$\end{document}FF¯), (b) mean R2 * (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{{{\rm{R}}}_{2}^{\ast }}$$\end{document}R2⁎¯) and (c) segmented volume (V), of the cervical-supraclavicular depot (sBAT) as obtained by the multi-atlas segmentation (MAS) method and the semi-automated reference segmentation (REF) method. Dashed lines indicate the 95% confidence interval for the regression line. Number of data points (subjects) = 16. The significance level was not adjusted for multiple comparisons.