Table 1. Segmentation results.
| Dataset Name | Init | SN15 | Melanoma | TScratch |
| Number of images | 28 | 54 | 20 | 24 |
| Patch Classifier Accuracy (%) | 95.5 | 94.5 | 90.5 | 89.8 |
| MultiCellSeg Accuracy (%) | 96.9 | 95.3 | 91.2 | 92.2 |
| TScratch Accuracy (%) | 92.3 | 92.3 | 87.0 | 89.8 |
| pValue: Patch Classifier vs. TScratch | 1.9e-4 | 1.8e-5 | 4.6e-5 | 0.95 |
| pValue: MultiCellSeg vs. TScratch | 9.1e-8 | 1.38e-5 | 3.37e-6 | 0.01 |
| Percent of Images for which MultiCellSeg Outperforms TScratch (%) | 95 | 85 | 90 | 75 |
Summary of segmentation accuracy and significance. Accuracy is defined as percent of correctly tagged pixels out of the total number of pixels in all images. Accuracy was calculated for the patches classifier (intermediate segmentation) and for the final MultiCellSeg segmentation and was compared to TScratch accuracy on the same set of images. pValue calculated as a paired t-test on the accuracy sequences: patches classification vs. TScratch MultiCellSeg vs. TScratch for each image. Percent of images for which MultiCellSeg outperforms TScratches' refers to the percent of images in the dataset that are better segmented by MultiCellSeg in comparison to TScratch.