Fig. 1From: Repeatability of two semi-automatic artificial intelligence approaches for tumor segmentation in PETTwo examples of a bounding box: A large bounding box is drawn around each lesion so that it also includes a large amount of background. For each lesion, the lesion is placed in a different position in the bounding box such that the CNN is not learning mainly the position of the lesion in the bounding box. On the left (a), the lesion is placed in the middle of the box, while on the right (b), the lesion is placed at the lower right borderBack to article page