search

baidu-research / NCRF

Cancer metastasis detection with neural conditional random field (NCRF)
Apache License 2.0
753 stars 184 forks source link

Can you explain how the code calculates the FROC score? #61

Closed alibalapour closed 1 year ago

alibalapour commented 2 years ago

I want to plot and calculate FROC for 3 WSIs (Test_001, Test_065, and Test_079 from Camelyon16), but the result is nan. I can't understand how do you calculate FROC. We have a CSV file that shows coordinates and probabilities of tumors generated by the model and ground truth mask. Is this ground truth mask have a particular property? What are the values of the ground truth mask? 0 or 1? What is ITC_labels in Evaluation_FROC.py? How do you generate a ground truth mask of WSI from annotations? Is there a particular way or the same as the way provided by ASAP?

tree1rain commented 1 year ago

Open tumor slide image with ASAP. Load the annotation file in .xml format. Save.

tree1rain commented 1 year ago

ASAP Save -> .tif -> assign labels to annotation groups -> (tumor: 1), (Exclusion: 0), (_0: 1), (_1: 1), (_2: 0) filled_image = nd.morphology.binary_fill_holes(binary) this line of code will fill in the hole, so there is no difference between (Exclusion: 0) and (Exclusion: 1) Evaluation_FROC.py -> computeEvaluationMask -> distance = nd.distance_transform_edt(255 - pixelarray[:,:,0]255)