Clarification on Microbleeds Inclusion in Lesion Count Metadata

[youwanM]

Hello,

Thank you for the effort put into creating such a useful dataset.

I noticed that in some subjects (such as sub-r040-s043) showing signs of microbleeds, the segmentation mask appears to include these microbleeds. I was wondering if the microbleeds are accounted for in the lesion count columns within the metadata.

Looking forward to your clarification.

Best regards,
Youwan

[leiliew]

Thank you so much for your comment, and apologies on our long delay. Could you send us an email to sliew@usc.edu and npnlusc@gmail.com with any subjects that you noticed this for? We will also complete a manual QC on our end, but if you specifically noticed numbers where this occured, it would help us greatly so we can fix this ahead of our next release. Thanks again so much!!!

[youwanM]

Hi @leiliew,

No problem, and thank you for the reply. This issue was originally noted by a visiting PhD student last year, but unfortunately, I’m not sure if the specific subject numbers were recorded.

In any case, this only affected a small minority of the images. From what I’ve found, 4 out of 655 subjects failed our quick QC. This includes the current case and the one mentioned in #13 . This leaves two other subjects for which I cannot recall the IDs or the specific reasons for their exclusion from the training set.

My apologies for not being able to retrieve those details.

Best,
Youwan