Fig. 12
From: Generalizable deep learning framework for 3D medical image segmentation using limited training data
Difference between model trained with 40 patients, "n=40 model", versus model trained using 3 patients, "n=3 model". Positive values show higher performance when trained with more training data