Fig. 5From: Combining generative modelling and semi-supervised domain adaptation for whole heart cardiovascular magnetic resonance angiography segmentationGAN-based method results on Dataset 1 (MMWHS). The first column shows the original input images, the following columns show outputs of the network for progressively higher level of supervision. Row 1–2 show generated images, while row 3 shows segmentation output. The GAN-based method transforms the source modality (CT) into target (MR) and vice versa. Legend for segmentation labels as follows: LV = turquoise, LVM = orange, RV = blue, LA = green, RA = yellow, AO = grey. GAN, generative adversarial networkBack to article page