Table 1.

Evaluation scheme of AI segmentations

Rating	Criteria
1. Perfect/optimal	The segmentation is as good as manual segmentation, that is, no false positive or false negative segmentations
2. Optimal with minor adjustments	The segmentation contains all lesions and only minor* false positives or false negatives
3. Acceptable with minor adjustments	The segmentation contains the majority of the lesions (at least 1 and ≤ 3 missing) and ≤ 2 false positive segmentations. Additionally only minor* false positives or false negatives
4. Acceptable with major adjustments	The segmentation contains most of the lesions (at least 1 and ≤ 6 missing) and ≤ 4 false positive segmentations. Additionally only minor* false positives or false negatives
5. Non-usable	The segmentation does not contain enough of the lesions (≥ 7 lesions missing or no lesions segmented if less than 7 lesions present) or too many false positives (≥ 5) for correction to be meaningful

* Minor is defined as only parts of a predicted lesion are wrong