kaggle-HuBMAP

HuBMAP - Hacking the Human Vasculature (Kaggle Competition)

Kidney microscopy resources

• kidney under a microscope
  • glomerulus: "little ball of yarn"
  • distal convoluted tubules: "open centers"
  • proximal convoluted tubules: "fluffy centers"

Image data overview

• training:

  • metadata summary csv: metadata/trn_img_meta_7033.csv
  • 7033 samples
  • 512 x 512 x 3
  • data type uint8
  • mean intensity across 3 channels:
  
  • stdev intensity across 3 channels:
  
  • sample image:
  

• test:

  • 1 sample
  • 512 x 512 x 3
  • data type uint8
  • mean 163.5
  • std 53.1
  • sample image:
  

Provided CSVs

• tile:
  • location: metadata/tile_meta_7033.csv
  • columns:
    • id
    • source_wsi (patients from where WSIs were extracted)
      • 1-4, 6-14
      • source_wsi 1-4 described in wsi_meta
    • dataset:
      1. 422
      2. 1211
      3. 5400
  • dataset description:
    1. Annotations that have been expert reviewed.
    2. Comprises the remaining tiles from these same WSIs and contain sparse annotations that have not been expert reviewed.
    3. Tiles extracted from an additional nine WSIs. These tiles have not been annotated. You may wish to apply semi- or self-supervised learning techniques on this data to support your predictions.
    • i, j: upper-left corner within the WSI where the tile was extracted
• wsi (whole slide images):
  • source_wsi (4 patients: 3 females, 1 male)
  • age
  • sex
  • race
  • height (cm)
  • weight (kg)
  • bmi

Annotations

• Source: polygons.jsonl
• 1633 unique IDs
• 3 labels:
  1. blood_vessel
  2. glomerulus
  3. unsure
• produced 1633 segmentations masks with name mask.tif
• sample image with segmentation: