def dice_4(y_true, y_pred):
    """
  Dice loss for class 4
  """
    batch_dice_loss = soft_sorensen_dice(y_true, y_pred, axis=[2, 3])
    mean_dice_loss = K.mean(batch_dice_loss, axis=0)
    return mean_dice_loss[4]
def dice_coef(y_true, y_pred):
    """
  Calculate DICE coefficient score for predicted mask. Expects true and
  predicted masks to be binary arrays [0/1]. 3D Volumetric DICE for one class.
  Take mean over batch
  
  # Arguments:
      y_true: numpy array of true targets, y_true.shape = [slices, H, W]
      y_pred: numpy array of predicted targets, y_pred.shape = [slices, H, W]
  # Returns:
      Scalar DICE coefficient for each label in the range [0 1]
  """
    batch_dice_loss = soft_sorensen_dice(y_true, y_pred, axis=[1, 2, 3, 4])
    mean_dice_loss = K.mean(batch_dice_loss, axis=0)
    return mean_dice_loss