Figure 2.

Boxplots showing the segmentation performances of in-house U-Net and open U-Net applied on the open and in-house test datasets. The methods are compared using the Dice similarity score (DSC). Mean (red dashed) and median (green) values are depicted. When applied to the in-house test dataset, the performance of the open U-Net generally drops and becomes more unreliable, which is depicted by a lower mean and median, as well as a larger spread between the best and worst DSC. Table 2 further gives an overview of the results.