EvaluateSegmentation Software

EvaluateSegmentation is a tool that compares two segmentations of volumes using 22 different metrics. These metrics were selected as a result of comprehensive research into the metrics used for evaluating medical volume segmentations. It is specifically optimised to be efficient and scalable, and hence can be used to compare segmentations on full body volumes. It can be downloaded from GitHub.

EvaluateSegmentation provides the following measures:

1. Dice Coefficient
2. Jaccard Coefficient
3. Area under ROC Curve (one system state)
4. Cohen Kappa
5. Rand Index
6. Adjusted Rand Index
7. Interclass Correlation
8. Volumetric Similarity Coefficient
9. Mutual Information

10. Hausdorff Distance
11. Average Distance
12. Mahanabolis Distance
13. Variation of Information
14. Global Consistency Error
15. Coefficient of Variation
16. Probabilistic Distance

Classic measures
17. Sensitivity (Recall, true positive rate)
18. Specificity (true negative rate)
19. Precision
20. F-Measure
21. Accuracy
22. Fallout (false positive rate)