DnaRnaProteins/ctc-cell-cycle-hela

CTC Cell Cycle Dataset

Cell Tracking Challenge (CTC) live-cell microscopy with derived cell cycle state labels for 3-class temporal classification.

What's actually hosted

The repo name says hela for historical reasons. Currently hosted: Fluo-N2DH-GOWT1 (GFP-tagged Oct4 in mouse embryonic stem cells), which is what the milestone baseline trained on. HeLa data may be added later under a hela/ prefix.

Sequence	Frames	Used as
01/	92	training
02/	92	held-out validation

Each sequence ships with the CTC ground-truth tracking under <seq>_GT/TRA/ (man_track*.tif masks + man_track.txt lineage) and segmentation under <seq>_GT/SEG/.

Label schema

Labels are derived algorithmically from lineage tree bifurcations — not annotated by hand.

ID	Label	Definition
0	interphase	Default proliferating state
1	pre-mitosis	Rounding phase within 8 frames before division
2	mitosis	Frame of division ±3 frames (lineage tree split)

Class distribution (seq 01): interphase ~98%, pre-mitosis ~1.2%, mitosis ~0.6%. Use class weights or balanced sampling.

Examples

samples/ contains rendered PNG previews:

• frame_full_*.png — full-frame examples at different timepoints
• crop_interphase.png, crop_premitosis.png, crop_mitosis.png — class examples

Loading

from huggingface_hub import snapshot_download
root = snapshot_download(repo_id="DnaRnaProteins/ctc-cell-cycle-hela", repo_type="dataset")
# Use with project/src/cell_cycle/datasets/ctc.py:CTCClipDataset

Source

MICCAI Cell Tracking Challenge Fluo-N2DH-GOWT1. Bartosova et al., FEBS Letters 2011 (original imaging); Maska et al., Nature Methods 2023; Ulman et al., Nature Methods 2017.