Fig. 4From: Fully-automated global and segmental strain analysis of DENSE cardiovascular magnetic resonance using deep learning for segmentation and phase unwrappingDemonstration of phase unwrapping of DENSE images (column a)Â using a semantic-segmentation U-Net, and comparison with the path-following method. For DENSE images with typical SNR (top two rows), the semantic segmentation U-Net correctly identified wrapped pixels (column b). Both the U-Net and the path-following methods performed phase unwrapping without errors (top two rows, columns c, d). For low-SNR data (bottom two rows), the U-Net successfully performed semantic segmentation and phase unwrapping, however the path-following method led to large phase-unwrapping errors, which are also depicted in wrap cycle maps (e)Back to article page