Figure 10
Pre-Annotation. The results of different pre-annotation methods (Otsu, percentile threshold, Cellpose39, U-Net40) are shown. Example images (first column) of the data introduced in Fig. 2 are illustrated compared to the annotation (second column). Only the U-Net has been trained on an already annotated sub-dataset. Here, 4% to 32% of all images in the dataset were annotated in advance. The other methods do not require an annotated sub-dataset. In addition to qualitative results in the form of contours or overlays, quantitative metrics (Dice Coefficient \(Q_{\text {DSC}}\) or advanced Aggregated Jaccard Index \(Q_{\text {AJI+}}\)) are presented.
