About human_expert label

[LinxuanHan]

Dear admin, sorry to bother you. The labels in csPCa_lesion_delineations/human_expert and csPCa_lesion_delineations/AI/Bosma22a have some issue, the content of the label file is zero both background and foreground. could you please answer the question? good luck.

[joeranbosma]

Hi @LinxuanHan,

For negative cases (approximately 2/3 of cases) the annotations are empty, indicating that case did not harbor clinically significant prostate cancer. For example case nr 32 had a clear tumor and therefore the label is not empty.

Hope this helps, Joeran

[LinxuanHan]

Hi Joeran， Thank you for your answer ,this helps me a lot. But I have other questions hope you can give me some tricks: 1、Can I use the AI labels to train a AI model aiming to segment ISUP≤1 and ISUP≥2 ？Because the labels were labeled by AI Model，I use it to deal with same thing，it looks like a bootstrap task。 2、The file in clinical_information/marksheet.csv ， about lines 12th，case_csPCA, does it mean the patient has csPCA and the csPCA is positive? (Table where "Yes" positive , "No" negetive ? or all the patients are csPCA positive but "Yes" table means ISUP≥2 “No" table means ISUP≤1) gool luck！

[joeranbosma]

Hi @LinxuanHan,

1. You can use the AI labels to train an AI model to segment ISUP ≥ 2 lesions (as indicated by the label map in the picai_labels repository). ISUP ≤ 1 lesions are not annotated (label 0) and are therefore indistinguishable from the background (also 0).
2. Having at least one csPCa lesion (ISUP ≥ 2) means the case is positive, and case_csPCa is "Yes". Lesions with ISUP 1 are not csPCa, as these indicate indolent cancer. So indeed, "Yes" means ISUP≥2 and “No" means ISUP≤1.

[LinxuanHan]

Hi Joeran

About last question 1 , AI labels have 3 content , I konw label 0 is background , but what do label 1 and label2 mean? src="https://github.com/DIAGNijmegen/picai_labels/assets/48483218/fd52223f-9755-4860-a1d1-18afd7bbcf1c">

[joeranbosma]

Hi @LinxuanHan,

The image you linked seems to be of a zonal prostate annotation! My answers above assumed you were asking about the lesion annotations (from csPCa_lesion_delineations).

Could you indicate from which folder the annotation is that you now have a question about? And what's the filename?

[LinxuanHan]

Hi Joeran, Sorry Joeran, it's my fault. The filename is \picai_labels-main\anatomical_delineations\zonal_pz_tz\AI\HeviAI23\10000_1000000.nii.gz, and I have already known why there are three kinds of labels ,according to your answer above. Because labels in zonal_pz_tz are about prostate zone , maybe Label 1 and Label 2 are pz and tz. So if I want to segment prostate and ISUP≥2，Can I use labels in anatomical_delineations\zonal_pz_tz\AI\Yuan23 and labels in csPCa_lesion_delineations\AI\Bosma22a at the same time ?

[joeranbosma]

Yes that's right!