SPIE-AAPM-NCI DAIR Digital Breast Tomosynthesis Lesion Detection Challenge (DBTex) - Phase 1 Forum

Go back to competition Back to thread list Post in this thread

> microcalcifications

Dear Organizers,

Your paper states that you created bounding boxes for masses and architectural distortions distortions, but there is no mention of microcalcifications. We were wondering how you handled them in your data set. Which of the following did you do?
1) Kept images containing them but did not create bounding boxes for them.
2) Discarded the images containing them.
3) Something else.

Posted by: nyu_bteam @ Jan. 12, 2021, 5:56 p.m.

Hi, apologies for the late response. Bounding boxes were not created for micro calcifications, but microcalcification cases were not eliminated. Some bounding boxes may contain micro calcifications, but they were not the reason for biopsy (reason for biopsy is a mass or an architectural distortion). Hope this helps!

Posted by: kdrukker @ Jan. 15, 2021, 4:26 p.m.

I find your answer a little puzzling.

As far as I know (and as I verified with a breast radiologist today),
1) microcalcifications are often biopsied, even when they are not coupled with a mass or an architectural distortion.

You are saying that:
2) There are no boxes for microcalcifications.
3) Cases with microcalcification were not eliminated.

The only way I can reconcile 1, 2 and 3 is by concluding that biopsied microcalfications don't have bounding box in the dataset, unless they were coupled with a mass or an architectural distortion. That is, that the task is to localize only the masses and architectural distortions that led to biopsies, while ignoring microcalcifications which led to biopsies. Is that correct?

Posted by: nyu_bteam @ Jan. 16, 2021, 5:06 a.m.

You are correct: the task is to detect masses and architectural distortions that led to biopsy.

Posted by: kdrukker @ Jan. 18, 2021, 5:01 p.m.

Ok, I understand that, but could you state whether the following statement is correct?
"The task is to localize only the masses and architectural distortions that led to biopsies, while ignoring microcalcifications which led to biopsies"

Posted by: nyu_bteam @ Jan. 18, 2021, 5:06 p.m.

Microcalcifications may or may not have led to biopsy and are not part of the detection task in this Challenge.

Posted by: kdrukker @ Jan. 18, 2021, 9:27 p.m.

Ok, so there are images in the dataset, which contain microcalcifications that led to a biopsy (which are not coupled with a mass or an architectural distortion), and these images will not have bounding boxes on these microcalcifications?

Posted by: nyu_bteam @ Jan. 18, 2021, 9:52 p.m.
Post in this thread