Table 7 Description of the validation metrics. Adapted from the work6.
From: Spinal cord gray matter segmentation using deep dilated convolutions
Metric Name | Abbr. | Range | Interpretation | Category |
|---|---|---|---|---|
Dice Similarity Coefficient | DSC | 0–1 | Similarity between masks | Overlap |
Jaccard Index | JI | 0–100 | Similarity between masks | Overlap |
Conformity Coefficient | CC | <100 | Ratio between mis-segmented and correctly segmented | Overlap |
Symmetric Mean Absolute Surface Distance | MSD | >0 | Mean euclidean distance between mask contours (mean error) | Distance |
Hausdorff Surface Distance | HSD | >0 | Longest euclidean distance between mask contours (absolute error) | Distance |
Skeletonized Hausdorff Distance | SHD | >0 | Indicator of maximal local error | Distance |
Skeletonized Median Distance | SMD | >0 | Indicator of global errors | Distance |
True Positive Rate or Sensitivity | TPR | 0–100 | Low values mean that method tends to under-segment | Statistical |
True Negative Rate or Specificity | TNR | 0–100 | Quality of segmented background | Statistical |
Positive Predictive Value or Precision | PPV | 0–100 | Low values mean that method tends to over-segment | Statistical |
