BioCluster

Provenance

Datasets

Public, deidentified, permissively licensed biomedical image datasets. Every field below is read from data/datasets.yaml, which is itself sourced from research/data_provenance.md.

PathMNIST (histopathology)

MedMNIST v2 (Yang et al.), derived from Kather et al. NCT-CRC-HE-100K + CRC-VAL-HE-7K

Modality
histopathology (H&E-stained colorectal tissue patches)
License
CC BY 4.0
Version
MedMNIST v2, Zenodo record 10519652
Sample unit
single non-overlapping tissue patch, native 28x28x3 (downsampled by MedMNIST from 224x224)
Patient/specimen unit
not individually resolvable in the public release
Site unit
resolvable: train/val = NCT-CRC-HE-100K, test = CRC-VAL-HE-7K (different clinical center)
Splitting unit
site (train/val pool vs. test pool are from different clinical centers)
Resolution
28x28x3 (native MedMNIST resolution used in this project; see research/decision_log.md)
Classes
9
Images (train/val/test)
89996 / 10004 / 7180
MD5
a8b06965200029087d5bd730944a56c1

Known biases

  • stain/scanning batch variation between centers
  • class imbalance (background/adipose overrepresented)
  • patch labels derived from whole-slide pathologist annotation, not independent per-patch review

Citation: See research/literature_matrix.csv rows for Kather et al. (2016/2018/2019) and Yang et al. MedMNIST v1/v2

BloodMNIST (blood cell microscopy)

MedMNIST v2 (Yang et al.), derived from Acevedo et al. peripheral blood cell dataset

Modality
blood cell microscopy (stained peripheral blood smear, single-cell crops)
License
CC BY 4.0
Version
MedMNIST v2, Zenodo record 10519652
Sample unit
single-cell crop, native 28x28x3 (downsampled by MedMNIST from a 200x200 center crop of a 360x363 source image)
Patient/specimen unit
not available — no donor ID field in the public release (see research/data_provenance.md)
Site unit
not available — no scanner/site metadata exposed
Splitting unit
image (random 7:1:2 split per MedMNIST documentation; NOT donor-stratified)
Resolution
28x28x3 (native MedMNIST resolution)
Classes
8
Images (train/val/test)
11959 / 1712 / 3421
MD5
7053d0359d879ad8a5505303e11de1dc

Known biases

  • class imbalance (platelets and immature granulocytes are minority classes)
  • single acquisition protocol per source paper, but this cannot be independently re-verified from the released images alone

Citation: See research/literature_matrix.csv row for Acevedo et al. and Yang et al. MedMNIST v1/v2

Patient/specimen-level leakage

PathMNIST's train/val pool (NCT-CRC-HE-100K) and test pool (CRC-VAL-HE-7K) come from different clinical centers -- a real, source-verified split used as the leakage boundary for every evaluation on that dataset. BloodMNIST exposes no donor identifier in its public release; independence there is assumed at the image level only, which is a documented limitation, not a guarantee.