Fig 1.
Still image (A) and mask (B), video screenshot (C) and mask (D), and 3D volume screenshot (E) and mask (F) from one patient.
Table 1.
U-Net training details.
Table 2.
Dataset features.
Table 3.
U-Net and NnU-Net 5-Fold Cross-Validation on TVUS uterus segmentation test set (Dice scores).
Table 4.
Optimal U-Net training, augmentation, and pre-processing configurations for each dataset.
Fig 2.
Visual results picked from the best, median, and worst model folds.
TVUS scans of the uterus manually segmented by a doctor (red) and overlayed by nnU-Net and U-Net model predictions (green) and their corresponding Dice scores.