Question about preprocessing the public EUS dataset into images/masks format

[tongdiedie]

Hi authors,

Thank you for releasing the TextSAM-EUS code.

I am trying to reproduce the experiments. I requested access to the public Endoscopic Ultrasound Database of the Pancreas, and the folder I received has this structure:

Ecoendoscopy Database/
  CANCER/
  HEALTHY PANCREAS/
  PANCREATITIS/
  CLINICAL INFORMATION.csv
  CLINICAL INFORMATION.xlsx

However, the TextSAM-EUS README expects:

data/EUS/
  train/images/
  train/masks/
  val/images/
  val/masks/
  test/images/
  test/masks/

The dataloader also appears to expect each image to have a corresponding same-name mask file.

Could you please clarify:

Where are the pixel-wise tumor masks located in the public database?
Is there a preprocessing/conversion script from the raw Ecoendoscopy Database folders to the images/ and masks/ structure?
Could you provide the patient-wise train/val/test split used in the paper?
Should the HEALTHY PANCREAS and PANCREATITIS folders be excluded for the tumor segmentation task, or used as all-background images?

This would be very helpful for reproducibility.

Thanks!