I use this module to compare segment and I use the same model. I don’t know why TruePositve value is less than TrueNegative value. I think this algorithm is measure from the box of pink. Am I true?? that the area outside the segment is TN and the segment model is a TP. I write on the picture down below.
I found this documentation source module and I think is it wrong measure? formula that dice similarity is use Union instead of intersect.
I surf on the internet that dice similarity look like this. Can anyone tell me that I got it right?
On a quick look, this seems OK to me. TP is less than TN because there are more voxels where the object isn’t present than where it is present. 20% of the voxels are object, and 80% are not object. Dice value is 1 because 2 * (0.2 * N) / (0.2N + 0.2N) = 1 (where N = number of voxels). The intersection of A and B is 0.2N because A has volume 0.2N and B has volume 0.2*N and all of those voxels are in the same place (because A=B) in your example.
1 Like
Thank you for your support.