metadata
license: cc-by-nc-nd-4.0
viewer: false
tags:
- pathology
- histopathology
- virtual-staining
- multiplex-immunofluorescence
task_categories:
- image-to-image
QUEST tutorial data
The three staged cohorts the QUEST tutorials read. Weights
live in yandrewl/QUEST; this repository is only the
data.
| cohort | tutorial | size | what it is |
|---|---|---|---|
stanford-pc/ |
1, 5 | 1.6 GB | 326 paired H&E / real-MIF patches over 29 regions, 42–52 markers each, plus the published per-patch cell composition |
bog-86337/ |
4 | 36 MB | one esophagus core: H&E mosaic, per-cell marker tables for its 16-region cohort on both the real and the predicted segmentation, and the frozen cell-state / neighbourhood vocabulary |
crc-metu/ |
6 | 73 MB | 62 colorectal slides from 57 patients, 37 events: per-patch virtual-MIF features, survival labels, and the fitted 5-fold out-of-fold MIL output |
from huggingface_hub import snapshot_download
snapshot_download("yandrewl/QUEST-tutorial-data", repo_type="dataset", local_dir="data",
allow_patterns=["crc-metu/*", "bog-86337/*", "stanford-pc/*"])
QUEST_DATA points elsewhere if you keep the cohorts outside the repo.
Tutorial 0 prints the shape
and dtype of every file here.
PathoCell (tutorials 2 and 3) is not staged: it is public and 36 GB whole, so
questkit.cohort.fetch_pathocell(["reg016_B"]) fetches only the regions a tutorial asks for from
Kainmueller-Lab/PathoCell.
Licence
CC BY-NC-ND 4.0, the same licence as the code and weights.