MedDPCR-Seg / README.md
DwD69's picture
Remove Attribution section from dataset card
bee1256 verified
|
Raw History Blame Contribute Delete
3.64 kB
metadata
pretty_name: MedDPCR-Seg
language:
  - en
task_categories:
  - image-segmentation
tags:
  - medical-imaging
  - text-guided-segmentation

MedDPCR-Seg

This collection brings together nine datasets for text-guided segmentation of breast lesions, brain tumors, and skin lesions. Each ZIP contains PNG images, segmentation masks, and spreadsheets with English text prompts.

Datasets and downloads

The split counts are the numbers of PNG images in each archive. UWaterlooSkinCancer has training and test images only.

Dataset Modality Anatomy Train Validation Test Download Original source
BUSI Ultrasound Breast 624 78 78 ZIP Al-Dhabyani et al.
BUSBRA Ultrasound Breast 1,311 282 282 ZIP Gomez-Flores, Gregorio-Calas and de Albuquerque Pereira
BUSUC Ultrasound Breast 567 122 122 ZIP Ahmed Iqbal; cite Iqbal and Sharif (2023)
BUID Ultrasound Breast 162 35 35 ZIP QAMEBI
UDIAT Ultrasound Breast 113 25 25 ZIP Dataset B
BTMRI MRI Brain 2,738 1,321 1,005 ZIP Jun Cheng
BRISC MRI Brain 4,000 1,000 1,000 ZIP BRISC dataset authors
ISIC Dermoscopy Skin 809 90 379 ZIP ISIC Challenge
UWaterlooSkinCancer Skin photography Skin 132 0 41 ZIP University of Waterloo

File layout

Extract each ZIP into a common data/ directory. The archive already contains its dataset folder. For example, extracting BUSI.zip creates the following layout:

data/BUSI/
    Train_Folder/
        img/
        label/
    Val_Folder/
        img/
        label/
    Test_Folder/
        img/
        label/
    Prompts_Folder/
        Train_text.xlsx
        Val_text.xlsx
        Test_text_original.xlsx
        ...

Using the text prompts

Choose the workbook for the desired split. Each row gives the image filename (Image), its English prompt (Description), and the mask filename (Ground Truth). Look up the image under img/ and the mask under label/ in that split. For example, the BUSI test workbook pairs Test_Folder/img/102_malignant.png with Test_Folder/label/102_malignant.png.

Use Test_text_original.xlsx for the standard test descriptions. Other test workbooks provide variants, including descriptions with the location omitted. Some datasets also include prompt lists for 10%, 25%, and 50% subsets of the training and validation splits.