Stand-alone MRI tool for semiautomatic volumetry of abdominal adipose compartments in patients with obesity

Figures

• 

  Fig. 1. Graphical user interface of stand-alone tool (segfatMR) for multi-slice segmentation of SAT and VAT amounts. After automatic segmentation, the user has the opportunity to review and graphically adjust the boundaries after selecting them by key press (1–3 for outer SAT, inner SAT and VAT, respectively). Unlike the SAT boundary, the VAT boundary is a crude one and fat quantification uses a histogram-based threshold (green dashed vertical line) to separate signal intensities from fat or lean tissues (resulting VAT areas overlaid in green). The interactive table summarizes all results and allows the user to link slices to positions of vertebral bodies (column VB) or landmarks (LM), such as pelvic floor (PF) or femoral heads (FH).

• 

  Fig. 2. Main processing steps for computation of refined outer SAT (a-d, left column), refined inner SAT (e-h, middle column) and crude (envelope) VAT boundaries (i-l, right column). (a) adaptive threshold mask of MR image using adaptthresh (Matlab function) with filter size 19 × 19. (b) binarized MR image using imbinarize and mask from a. Resulting white areas primarily reflect adipose tissue but artefacts are present as well, most notably, at the bottom of the image. (c) largest connected white-pixel region from b determined using regionprops is considered as SAT. Further criteria were implemented for consistency, for example, the ratio of the principal diameters of considered regions. (d) resulting outer (yellow) SAT contour computed using bwboundaries on filled mask (not shown here) and overlaid on original MR image. (e) adaptive threshold mask with 17 × 17 filter used for binarization and (f) resulting binary mask (inverted). (g) largest connected region of mask f after multiplication with filled mask for outer SAT contour. (h) resulting inner (blue) SAT contour. (i) filled mask for inner SAT contour after evaluation of different threshold masks and quantitative criteria (region…

• 

  Fig. 3. Linear regression results (a,b) and Bland-Altman plots (c,d) of abdominopelvic SAT volumes (20 patients) segmented with new (segfatMR) and reference method (sliceOmatic) for two independent readers (R1: a, c and R2;b,d).

• 

  Fig. 4. Linear regression results (a,b) and Bland-Altman plots (c,d) of total VAT volumes (20 patients) segmented with new (segfatMR) and reference method (sliceOmatic) for two independent readers (R1:a,c and R2;b,d).

• 

  Fig. 5. Linear regression results (a,b) and Bland-Altman plots (c,d) of abdominopelvic SAT volumes (20 patients) segmented by two readers (R1 and R2) for two methods (ref: a,c and new b,d).

• 

  Fig. 6. Linear regression results (a,b) and Bland-Altman plots (c,d) of total VAT volumes (20 patients) segmented by two readers (R1 and R2) for two methods (ref:a,c and new b,d).

• 

  Fig. 7. Boxplot of effective processing times (in seconds) per slice for fat segmentation with the new (fatSeg) and the reference method (sliceOmatic) and both readers (R1 and R2). Boxes mark quartiles Q1-Q3 and whiskers minimum and maximum times.