Volume similarity has the wrong sign about segmentation_metrics HOT 3 CLOSED

@jhlegarreta Thank you for pointing it out! I am checking and fixing this issue.

from segmentation_metrics.

@jhlegarreta Thank you for your advice again. I updated the package and now you can get the correct volume similarity sign by updating the package to the latest version (>=1.2.8)
pip install --upgrade seg-metrics

Details of update: I removed the old code dicecomputer.Execute(labelPred > 0.5, labelTrue > 0.5) because it relied on ITK which was not explicit enough (the document of ITK for Python is not complete as well). I added the following code to explicitly calculate VS (also added the check of extreme conditions):

v_pred = np.sum(labelPred > 0.5)
v_gdth = np.sum(labelTrue > 0.5)
if v_pred + v_gdth == 0: # prediction results and ground truth are all 0
    vs = 0
else:
    vs = 2 * (v_pred - v_gdth) / (v_pred + v_gdth)

If you have more advice, please let me know.

from segmentation_metrics.

@Jingnan-Jia sounds good. Thanks for the update. Closing the issue as resolved after bc37480.

from segmentation_metrics.