Skip to main content
. Author manuscript; available in PMC: 2020 Aug 31.
Published in final edited form as: Proc IEEE Int Symp Biomed Imaging. 2019 Jul 11;2019:348–351. doi: 10.1109/isbi.2019.8759295

Fig. 3.

Fig. 3.

Intra-site and inter-site testing illustrates the importance of image normalization methods on prostate gland segmentation performance (Dice overlap) when training with data from a single site. We show results for models trained using Stanford images alone and tested on data from either Stanford or Yale, as well as results for models trained using Yale images alone and tested on data from either Stanford or Yale. Boxplots show the median, 25th and 75th percentiles, extremes (approximately the middle 99.3%), and outliers.