Table 1.
Our updated nomenclature, expanding on the work of Nascimento and Marques133 which focused on the comparison between manual and automated segmentations (Algorithm vs. Manual), to also cover the case when two manual segmentations are being compared (Manual vs. Manual).
| Match Type | Manual vs. Manual | Algorithm vs. Manual |
|---|---|---|
| 1-1 | Expert Agreement | Correct Detection |
| 1-N | Ambiguous Masks | Merge |
| M-1 | Ambiguous Masks | Split |
| M-N | Ambiguous Masks | Split-Merge |
| 0-1 | Expert Disagreement | Detection Failure |
| 1-0 | Expert Disagreement | False Alarm |
Examples of the different classes, for the situation of Algorithm vs. Manual, are shown in Fig. 1. The top four classes (1-1, 1-N, M-1, and M-N) represent cases of segmentation agreement, though the number of lesions and the boundary of those lesions is disputed. Whereas the bottom two classes (1-0, 0-1) are the classes which summarize segmentation disagreement.