Request for Dataset Processing Scripts

[TanmDL]

Hello,

Your work is fantastic, and I truly appreciate the effort that went into it. However, I have a few questions about the processing of the two datasets. If you could share the scripts used for this, it would be very helpful for us. I assure you that I will properly cite your project.

Thank you!

[AmayaGS]

Hi @TanmDL, really sorry it took me a while to answer you! Thanks very much for your interest in this work :)

All the scripts I used for the preprocessing are on the GitHub repo. I just updated the README file with instructions for how to run the code, including the preprocessing steps - including the tissue segmentation, patching and feature embedding. Maybe this might answer your questions, but if not please let me know which specific step needs further explanation. Hope this helps!

[AmayaGS]

Any thoughts you might have to make the README clearer for users would be very valuable to me, so please let me know if there's anything unclear/which could be explained in more detail :)

[TanmDL]

Thank you for your response. Could you please clarify which classification task was studied here? Initially, I thought it would involve classes like CD68 and CD138, but I found the following labels: label_dict: {'0': 'Pauci-Immune', '1': 'Lymphoid/Myeloid'}, which appear to represent subtypes. This has confused me. Could you please explain this part in more detail? Also, I want to add that the pipeline design is excellent. Thank you.

[AmayaGS]

For the Rheumatoid Arthritis dataset I classified into inflammatory subtypes {'0': 'Pauci-Immune', '1': 'Lymphoid/Myeloid'} and for the Sjogren dataset into Absence and Presence of of Sjogren {'0': 'Not Sjogren', '1': 'Sjogren'}. However, depending on your data structure I think you could use the code to target your stains as labels. If you give me more detail on that, I could suggest how to do it. For example, assuming you have multiple stains per patient and want to classify the stains, you could add a column to patient_labels.csv file like so:

Patient_ID	Patient_stains	Patient_stains_numeric	label
Patient1	Patient1_CD68	Patient1.1_CD68	1
Patient1	Patient1_CD138	Patient1.2_CD138	2
Patient1	Patient1_CD20	Patient1.3_CD20	3
Patient1	Patient1_CD21	Patient1.4_CD21	4
Patient2	Patient2_CD68	Patient2.1_CD68	1
Patient2	Patient2_CD20	Patient2.3_CD20	3
Patient2	Patient2_CD21	Patient2.4_CD21	4

# Label/split configurations
labels:
  label: 'label' # column name for target label
  label_dict:  {'CD68': 1, 'CD138': 2, 'CD20': 3, 'CD21': 4} # Stain type numeric coding dictionary
  n_classes: 4 # number of target classes
  patient_id: 'Patient_stains_numeric' # column name for each unique file 

# Parsing configurations 
parsing:
  patient_ID: 'img.split("_")[0]' # "Patient1.1_stain" -> Patient1.1
  stain: 'img.split("_")[1]' # "Patient1.1_stain" -> stain
  stain_types: {'NA': 0, 'CD68': 1, 'CD138': 2, 'CD20': 3, 'CD21': 4} # Stain types 

Of course this off the top of my head - I haven't tested it and it would depend on your file structure, but it should work.

[TanmDL]

Thank you for your kind reply and giving me a fantastic idea. Can you please send me dataset links so that I can try to download and test them?

[AmayaGS]

Unfortunately, for patient privacy protection, I am not able to share these datasets publicly as they come from clinical trial and research datasets. I am currently exploring options to publish a multistain dataset and can let you know how that goes, however if it works out it wouldn't be until mid-next year. Very sorry about that!