Table 2 Dice similarity coefficient (DSC) for all model configurations. Symbol ↑ indicates that higher values are better. The best values from each measurement are written in bold and and second-best values are underlined. As can be appreciated, in this evaluation, the PUNet-wSL models performed better in segmenting changing WMH (i.e., the combination of ‘Shrinking’ and ‘Growing’ WMH) than the original U-Net, which performed better on segmenting ‘Stable’ WMH. PUNet-wSL-vol had an overall better performance in estimating the future volume of WMH, as per Table 1.

From: Prediction of white matter hyperintensities evolution one-year post-stroke from a single-point brain MRI and stroke lesions information