VNET Dice loss higher than 1

Question

I am training a Vnet on a segmentation problem with Keras 2.2.4 and Tensorflow 1.12.0 (I cannot change the tensorflow version). The masks, i.e. y_true, are (1,200,150,100,2)