Figure 4.

Comparison of DeepLabv3-based and SegFormer-based structure-specific and combined segmentation model performance on an external laparoscopic image dataset (LapGyn4). Models were deployed to the publicly available LapGyn4 dataset of non-semantically segmented images from gynecological procedures in conventional laparoscopic technique. Model segmentations for each organ are displayed. For the structure-specific models, segmentations of the 11 individual segmentation models are overlayed in one image. The figure shows representative images from the dataset.