Fig. 10: Ablation studies of augmentation strategies, architectural components, and parameter sensitivity in GenSeg.

a (Left) Impact of augmentation operations on the performance of GenSeg-UNet was evaluated on the test datasets of JSRT, NLM-MC, and NLM-SZ, in lung segmentation. GenSeg-UNet was trained using 9 examples from the JSRT training dataset. ALL refers to the full GenSeg method that incorporates all three operations. (Right) Impact of augmentation operations on the performance of GenSeg-UNet was evaluated on the test datasets of ISIC, PH2, and DermIS, in skin lesion segmentation. GenSeg-UNet was trained using 40 examples from the ISIC training dataset. b, c Ablation study evaluating the impact of elastic augmentation under in-domain (b) and out-of-domain settings (c). In out-of-domain scenarios, datasets are denoted in the format X-Y, where X represents the training dataset and Y the test dataset. UNet was used as the segmentation model. d Ablation study evaluating the impact of rotation augmentation on placental vessel segmentation using the FetReg and FPD datasets with UNet as the segmentation model. e Ablation study on learnable multi-branch convolutions, with UNet as the segmentation model. f (Left) Impact of the tradeoff parameter γ on the performance of GenSeg-UNet on the test datasets of JSRT, NLM-MC, and NLM-SZ, in lung segmentation with 9 examples from the JSRT training dataset. (Right) Impact of the tradeoff parameter γ on the performance of GenSeg-UNet on the test datasets of ISIC, PH2, and DermIS, in skin lesion segmentation with 40 examples from the ISIC training dataset. In all panels (except f), bar heights represent the mean, and error bars indicate the standard deviation across three independent runs with different random seeds. Results from individual runs are shown as dot points. Source data are provided as a Source Data file.