The authors presented a semi-automatic live tumor segmentation technique. In this method, the liver is first segmented using watershed algorithm, tumors are then segmented using a minimum cross-entropy multi-thresholding algorithm. The tumor segmentation is further refined with region growing and level-set based surface smoothing methods.

Detailed comments:

1) Overall, this is a nice workshop paper. The presentation is clear.

2) In the multi-thresholding tumor segmentation, the number of thresholds applied in multi-thresholding is 3, with the first and third thresholds set for hypodense and hyperdense tumors respectively. What is the second threshold for?

3) In segmentation refinement, why are the threshold limits not symmetric with respect to the mean value?

4) Could the user manually identified point in the initialization be an arbitrary point in the liver, e.g. a point in normal liver tissue or in the lesion?