Table 1.
Clinically relevant quantitative parameters for describing VH.
Fig 1.
Overview of the anatomical labeling protocol.
(a) Axial and sagittal slices to label are determined in terms of the size and resolution of the volume. (b) On the selected axial slices, the anterior (outer and inner borders) and posterior abdominal wall is traced. At the same time, linea alba and linea semilunaris are labeled on the appropriate axial slices. (c) The VH is labeled entirely on every axial slice where the hernia exists. (d) On the selected sagittal slices, the outer and inner borders of the anterior abdominal wall are traced. Note the previous VH and abdominal wall labels can be helpful references. (e) The umbilicus and skeletal landmarks are labeled. (f) The complete set of labels is reviewed.
Fig 2.
Anatomical structures included in the CT labeling protocol.
(a) rectus muscles; (b) oblique abdominal muscles; (c) linea alba; (d) linea semilunaris; (e) umbilicus; (f) xiphoid process; (g) anterior superior iliac spines; and (h) pubic symphysis.
Fig 3.
Examples of various ventral hernia sizes.
(a), (b), (c) demonstrate a small, medium, and large hernia, respectively in axial slices. The herniated regions are highlighted in red.
Table 2.
Quantitative evaluations on 20 derived metrics.
Table 3.
Abdominal wall reliability measured by mean surfaces distances (MSD) and Hausdorff distance (HD) in mm.
Table 4.
Fascial boundaries and bony structures reliability measured by Euclidean distance (ED) of centroids in mm.
Table 5.
Hernia volume reliability.
Table 6.
Statistical comparison of 20 metrics between two groups of patients with distinct outcomes.
Fig 4.
Results of preliminary statistical analyses.
(a) and (b) shows the number of false predictions and number of included variables over different alpha values using cross-validated elastic net regularized logistic regression, respectively. Generally, a larger alpha value yields stronger regularization, and thus involves less variables for the regression model. Note that the blue dashed curves represent the regression results using EHSCHV variables, while the green solid curves use the variables derived from labeling. (c) presents a hyper-plane using support vector machine to separate the two groups of patients with distinct technical outcomes by the two remaining labeling-derived variables of an exploratory regression model built upon all observations.
Fig 5.
Two VH cases in volume rendering and tri-planar views.
Although the two examples have almost the same hernia volume size (a = 125 cm3, b = 109 cm3), (a) is a long, shallow rupture at the umbilicus, while (b) is a short, deep protrusion of the abdominal wall. In addition, the patients’ body sizes are quite different, and the hernia in (b) is further away from the umbilicus.
Fig 6.
Illustration of VH characteristics on CT for four patients.
In each section, the first row illustrates the location of the VH; the second row illustrates the VH defect size at the anterior abdominal wall; the third row demonstrates the volume size of the hernia sac (red) and the abdominal cavity (blue).
Fig 7.
Illustration of VH characteristics in terms of processed label results.
The first row, from (a) to (c), demonstrates a matchup between the original image data and the processed labels, where the abdominal walls were interpolated. The second row, from (d) to (f), demonstrates the coherence of interpolated abdominal walls with the original image in three different views. The third row, from (g) to (i), illustrates a combined model of abdominal wall and hernia volume for shape-related VH characteristics, the relative location of VH with respect to the linea alba and linea semilunaris, and the relative location of VH with respect to skeletal landmarks and the umbilicus. The fourth row, from (j) to (l), demonstrates feasibility of measuring the VH defect size, width and length of VH, and ratio of volume size between the hernia sac and the abdominal cavity.