cassie07 / pathomics Goto Github PK

Hi! Thanks for your work. But I cannot find the preprocessing, e.g., the patching code, and I don't know which magnification you use in your work. Can you share the preprocessing codes to increase reproducibility?

I try to run the bash_main.py on COAD in different CUDA enviroment. But, I encounter the same cuda error.

../aten/src/ATen/native/cuda/ScatterGatherKernel.cu:144: operator(): block: [0,0,0], thread: [0,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "index out of bounds" failed.

1. How does the WSI patching process work?
2. Facing the problem of missing npy files.
   This is probably because there is no image feature extraction process.
   Do you have a script for image feature extraction? Or can you tell me how to proceed?

Thanks for doing the research.

Thanks for your grate work! In bash_main.py, default feature_dimension is 1024. However, resnet 50 and 101 generates feature dimension 2048; resnet 18 generates feature dimension 512. How to get feature dimension 1024?

Thank you for this research! I am a beginner and I have a question to ask. Why do we need to group image features during pre-training? The grouping is integrated into eight 1×256 vectors. Finally, global attention is used to synthesize a 1×256 vector. Why not do it directly on the data before grouping, and directly add some layers of the network after the K×1024 vector extracted from the original WSI and get a 1×256 vector?

感谢你们的这项研究！我是一个初学者，有一个问题想请教一下，请问在预训练中为什么要对图像特征分组呢？分组整合为8个1×256的向量最后还要用全局注意力合成一个1×256的向量，为什么不直接对分组前的数据做，直接把大小为K×1024的数据提取出来的向量通过加几层网络变成1×256的向量呢？

1.how can we get the floder ‘TCGA-COAD/Extracted_feature' ?
2.how can we generate npy files?

In your code, it seems that the genetic data has a certain format,
df_cna = pd.read_csv(z_score_path, delimiter = "\t")
df_cna = df_cna.drop(['Entrez_Gene_Id'], axis=1)
df_cna = df_cna[df_cna['Hugo_Symbol'].notna()].dropna()
df_cna = df_cna.set_index('Hugo_Symbol')
df_gene = df_cna
Can you tell me what format the different genetic data have besides the gene name?

Hi, I have a problem regarding stacking the gene data. In your code when we collect x_omic they collect genes from gene_family_dict and they can be in different sizes, so at the end, we have a list of tensors with different sizes. When we want to make the dataloader and in collate_fn how do you manage that to stack tensors with different sizes?

It seems that you have made adjustments to the clinical data. Can you tell me what the adjustments are?