Skip to content

Document official PhysioNet VinDr-CXR files in VinDr_Dataset docstring - #196

Open
AmlanMishra2004 wants to merge 1 commit into
mlmed:mainfrom
AmlanMishra2004:vindr-physionet-docs
Open

AmlanMishra2004 wants to merge 1 commit into
mlmed:mainfrom
AmlanMishra2004:vindr-physionet-docs

Conversation

@AmlanMishra2004

Copy link
Copy Markdown
Contributor

Docs-only change to the VinDr_Dataset docstring.

What

  • Notes that the bundled default CSV (vinbigdata-train.csv.gz) is the Kaggle challenge training set: 15,000 images, each read independently by 3 radiologists.
  • Explains that the official PhysioNet release (https://physionet.org/content/vindr-cxr/1.0.0/, credentialed access) also includes a 3,000-image test set labelled by the consensus of 5 radiologists, and how to load it: point csvpath at annotations/annotations_train.csv or annotations/annotations_test.csv and imgpath at the matching train or test folder. Only the classes the loader already lists are read; the extra local labels in those files (e.g. Rib fracture, Emphysema) are ignored.
  • Adds the PhysioNet link alongside the Kaggle challenge link.
  • Fixes the source hospitals: the data came from Hanoi Medical University Hospital and Hospital 108 (per the paper and the PhysioNet page), not Bach Mai Hospital.

The PhysioNet files are not bundled, since they are covered by the PhysioNet Credentialed Health Data Use Agreement.

Checked

With the current loader, VinDr_Dataset(imgpath=..., csvpath=<official annotations CSV>) loads 15,000 images from annotations_train.csv and 3,000 from annotations_test.csv, and the PhysioNet folder layout (train/<id>.dicom, test/<id>.dicom) matches what __getitem__ expects.

Independent of #195 (no overlapping lines; they merge cleanly in either order).

Made with Cursor

Explain that the bundled CSV is the Kaggle training set and how to load the
official PhysioNet annotations, including the consensus-labelled test set.
Also correct the source hospital (Hospital 108, not Bach Mai).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant