search

lsqshr / AH-Net

The Pytorch implementation of the 3D Anisotropic Hybrid Network described in the paper "3D Anisotropic Hybrid Network: Transferring Convolutional Features from 2D Images to 3D Anisotropic Volumes"
https://arxiv.org/abs/1711.08580
53 stars 11 forks source link

Metrics for TP/FP detection #3

Open Westerby opened 5 years ago

Westerby commented 5 years ago

Hello, in the paper it is mentioned that "true positive finding if the maximal point resides in a 3D bounding box annotated by the radiologist". Do you know if that was dealt with automatically? How were predictions assigned to GT boxes if there were several detections and several bounding boxes?

Thanks.

lsqshr commented 5 years ago

Not sure if I understood your question. True positive here means true positive markers that can be catched by a lesion finding annotated by the radiologist (3D bounding box).

There could be multiple detected markers outputted by the nonmaximal suppression in the same volume. The maximal point here means the local maximals. Not the global maximal.

Hope I answered your question.

For a good example to compute the FROC curve, you can check the script from the LUNA challenge:https://luna16.grand-challenge.org/evaluation/

On Tue, Oct 8, 2019, 6:04 AM Westerby notifications@github.com wrote:

Hello, in the paper it is mentioned that "true positive finding if the maximal point resides in a 3D bounding box annotated by the radiologist". Do you know if that was dealt with automatically? How were predictions assigned to GT boxes if there were several detections and several bounding boxes?

Thanks.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/lsqshr/AH-Net/issues/3?email_source=notifications&email_token=AAL6ZBJ7LVCDCICBVWGJKXTQNRLLTA5CNFSM4I6PUT5KYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HQJVPNA, or mute the thread https://github.com/notifications/unsubscribe-auth/AAL6ZBO3PSG7FRTBOKTVEITQNRLLTANCNFSM4I6PUT5A .