The anatomical variable in the slide level feature extraction code

[Zhcyoung]

full_path = r'./Downstream/Tumor_origin/src/feature/tcga/TCGA-LN-A8I1-01Z-00-DX1.F2C4FBC3-1FFA-45E9-9483-C3F1B2B7EF2D.pt' 
features = torch.load(full_path, map_location=torch.device('cpu'))
anatomical = 13
with torch.no_grad():
    x, tmp_z = features, anatomical
    result = model(x, torch.tensor([tmp_z]))
    wsi_feature_emb = result['WSI_feature']  ###[1,768]
    print(wsi_feature_emb.size())

It is unclear what the variable anatomical represents. The value 13 is assigned to it, but there is no explanation about its meaning.

Questions:

1. Is anatomical a specific value that needs to be set based on certain criteria, or can it be arbitrarily chosen?
2. If it is not arbitrary, how should I determine the correct value for anatomical?

Could you please provide clarification on what anatomical represents and how to determine the correct value for it in this context?

Thank you!

[Zhcyoung]

Oh, I found the meaning of anatomical!

[SeaTalk]

what does it mean?😳😳

[Zhcyoung]

what does it mean?😳😳

The anatomical variable and its code can be found here: CHIEF/configs/anatomic_mapping.yaml
Anatomic: 'brain':0, 'breast':1, 'bladder':2, 'kidney':3, 'prostate':4, 'testis':5, 'lung':6, 'pancreas':7, 'liver':8, 'skin':9, 'ovary':10, 'cervix':11, 'uterus':12, 'colon':13, 'esophagus':14, 'stomach':15, 'thyroid':16, 'adrenal gland':17, 'soft tissue':18

[SeaTalk]

what does it mean?😳😳

The anatomical variable and its code can be found here: CHIEF/configs/anatomic_mapping.yaml Anatomic: 'brain':0, 'breast':1, 'bladder':2, 'kidney':3, 'prostate':4, 'testis':5, 'lung':6, 'pancreas':7, 'liver':8, 'skin':9, 'ovary':10, 'cervix':11, 'uterus':12, 'colon':13, 'esophagus':14, 'stomach':15, 'thyroid':16, 'adrenal gland':17, 'soft tissue':18

Thank you very much! By the way, do you know how to generate .pt files like /Downstream/Tumor_origin/src/feature/tcga in which each file size is up to like a few decades of MB?

That confused me for a while~ Hope to find some help!

XIEXIE!

[Zhcyoung]

I think it is derived from the patch features:
image = Image.open("./exsample/exsample.tif")
image = trnsfrms_val(image).unsqueeze(dim=0)
with torch.no_grad():
patch_feature_emb = model(image) # Extracted features (torch.Tensor) with shape [1,768]
print(patch_feature_emb.size())