Table 1.

Main characteristics of the models studied.

Model	Category	Main characteristics
nnFormer43	Hybrid	Local and global attention blocks, convolutionnal down-sampling
nnUNet33	CNN	U-Net featuring five encoder and decoder blocks
SegmentationNet44	CNN	U-Net featuring four encoder-decoder blocks
Swin-UNetr45,46	Hybrid	Swin-Unet47 based encoder, CNN decoder, pre-trained with SSL
Transbts48	Hybrid	Large transformer bottleneck, CNN encoder/decoder
UNetr49	Hybrid	Transformer based encoder, CNN decoder
VT-UNet50	Transformer	Alternates regular and shifted attention windows layer, pre-trained with Swin T14