[clinical] Fix naming of the tables associated with analysis results

[fedorov]

See related discussion in https://nciimagingdat-m8a6349.slack.com/archives/C045VQUAZ0S/p1734563139760659.

The convention introduced in v20 of assigning names as <collection_id>_bamf_<cancer_location>_<modality>_segmentation should be reverted.

I suggest for analysis results we assign the name as <analysis_results_collection_id> (if there is no name assigned to the individual table), and analysis_results_collection_id column to the columns_metadata table.

[bcli4d]

An issue with your proposal is that, for both liver and lung cancer collections, there are two csv files. E.G. for lung cancer there is a file corresponding to CT segs and a file corresponding to FDG PET/CT segs. Similarly for liver cancer...CT and MR

So we'd have to merge these files , and, in the case of lung cancer, the files have slightly different columns: the CT seg file has a 'SeriesInstanceUID' column. The FDG PET/CT file has PTSeriesInstanceUID and CTSeriesInstanceUID columns.

Also, probably need to add a column to differentiate modalities?

[fedorov]

As with any collection, we will need to have custom rules how to assign the suffix of the table. I think those rules should follow the conventions established by the creators of those tables - not by us. This way it will be easier for the users who look at the collection at the page pointed by DOI to reconcile organization of the accompanying file with respect to what we have in BQ.

As another general rule, we do not harmonize and do not modify the tables we ingest, other than making sure PatientID in the images matches dicom_patient_id field in the ingested tables.

I didn't look at the specifics for this collection, but doing this now, from https://zenodo.org/records/13244892, there is a bunch of zip files, and (I assume - I did not look at every zip file) each zip file contains file named qa-results.csv that we need to ingest. I suggest we could use the pattern BAMF_AIMI_Annotations_<zip file prefix>_qa_results, such as BAMF_AIMI_Annotations_lung_fdg_pet_ct_qa_results.

This also reminds me that we should revisit rules for assigning IDs to the analysis results - it is quite confusing that collection_id and analysis_results_collection_id follow different conventions.

[bcli4d]

Which specific conventions are not consistent? There are potentially several.

[bcli4d]

The original_collections_metadata table has:

• collection_name: Collection name as used externally by IDC webapp
• collection_id: Collection ID as used internally by IDC webapp
  and analysis_results_metadata has:
• ID: Results ID
• Title: Descriptive title

We could go to:

• collection_name: Collection name as used externally by IDC webapp
• collection_id: Collection ID as used to identify collections in other BQ tables
• collection_title: Descriptive title
  and
• analysis_name: Analysis name as used externally by IDC webapp
• analysis_id: Analysis ID as used to identify analysis results in other BQ tables
• analysis_title: Descriptive title

So, for BAMF:
analysis_name: BAMF_AIMI_Annotations_lung_fdg_pet_ct_qa_results
analysis_id: bamf_aimi_annotations_lung_fdg_pet_ct_qa_results
analysis_title: Image segmentations produced by BAMF under the AIMI Annotations initiative

Then, instead of BAMF_AIMI_Annotations_lung_fdg_pet_ct_qa_results, we'd go to bamf_aimi_annotations_lung_fdg_pet_ct_qa_results

[bcli4d]

I'm working on restructuring BAMF file names...
The table_metadata clinical BQ table has one or more rows per collection where each row describes a table which contains data of patients in that collection. Should each bamf_aimi_table be associated with each collection that has patients in the bamf_aimi_table? Or, should bamf_aimi_annotations be in the collection_id column?
E.G., currently table_metadata has this partial row:

anti_pd_1_lung	bigquery-public-data.idc_v20_clinical.anti_pd_1_lung_bamf_lung_ct_segmentation	segmentation of Lung and Nodules (3mm-30mm) from CT scans

Should this now be replaced by:

anti_pd_1_lung	bigquery-public-data.idc_v20_clinical.bamf_aimi_annotations_lung_ct_qa_results	segmentation of Lung and Nodules (3mm-30mm) from CT scans

and with a similar row for the other collections, lung_pet_ct_dx, etc., that have patients in bamf_aimi_annotations_lung_ct_qa_results? Or should those rows be deleted, and instead have a row for each bamf_aimi_annotations table:

bamf_aimi_annotations	bigquery-public-data.idc_v20_clinical.bamf_aimi_annotations_lung_ct_qa_results	segmentation of Lung and Nodules (3mm-30mm) from CT scans
bamf_aimi_annotations	bigquery-public-data.idc_v20_clinical.bamf_aimi_annotations_lung_fdg_pet_ct_qa_results	segmentation of Lungs and FDG-avid lesions in the lung from FDG PET/CT scans
...

[bcli4d]

Please look at idc-dev-etl.idc_v21_clinical. So, there are now 11 BQ tables named like bamf_ami_annotations_*_qa_results.
I implemented the first approach above. Specifically, the table_metadata BQ table does not include bamf_aimi_annotations in the collection_id column.

[bcli4d]

Also, please look at my note above on conventions for original_collections_metadata and analysis_results_metadata

[fedorov]

@bcli4d yes, this looks good to me - thank you!

This also reminds me that we should revisit rules for assigning IDs to the analysis results - it is quite confusing that collection_id and analysis_results_collection_id follow different conventions.

Which specific conventions are not consistent?

I referred to this: collection_id are lowercase + underscore, analysis_results_collection_id is lower/uppercase + dashes.

I know this is the convention we inherited from TCIA.

[bcli4d]

Currently, original_collections_metadata has:

• collection_name: Collection ID as used externally by IDC webapp
  • Mixed case
  • Mixture of "-", "_" and " "
• collection_id: Collection ID as used internally by IDC webapp
  • Lower case
  • "_"

and analysis_results_metadata has:

• ID: Short ID
  • Mixed case
  • "-"
• Title: Descriptive title
  • Arbitrary text

We could go with:

• collection_name: Collection ID as used externally by IDC webapp
  • Mixed case
  • Mixture of "-", "_" and " "
• collection_id: Collection ID as used internally by IDC webapp
  • Lower case
  • "_"
• collection_title: Descriptive title
  • Arbitrary text

and analysis_results_metadata has:

• analysis_name: Analysis ID as used externally by IDC webapp
  • Mixed case
  • "-"
• analysis_id: Analysis ID as used internally by IDC webapp
  • Lower case
  • "_"
• analysis_title: Descriptive title
  • Arbitrary text

Then for collection CCDI-MCI, we'd have:
collection_name: CCDI-MCI
collection_id: ccdi_mci
collection_title: CCDI-MCI: DICOM converted whole slide hematoxylin and eosin stained images from the Molecular Characterization Initiative of the National Cancer Institute's Childhood Cancer Data Initiative

and for analysis result BAMF-AIMI-Annotations:
analysis_name: BAMF-AIMI-Annotations
analysis_id: bamf_aimi_annotations
analysis_title: BAMF-AIMI-Annotations: Image segmentations produced by the AIMI Annotations initiative

[fedorov]

Yes, we could, but this would introduce yet another "flavor", since TCIA analysis results collections would have a different conventions for the analysis results IDs. How about we bring this up in IRCoCo?