fidoptZephIR worm (freely moving C. elegans, whole-brain imaging)
Two volumes of one fluorescence recording of a freely moving C. elegans head, 16 to 123 seconds apart, with the 178 tracked neurons given a point in both, placed or checked by people (ZephIR).
Read before comparing (10)
- Recording. 1,060 volumes at 4 Hz of a ZM9624 worm crawling freely, which the provider centred and rotated rigidly but did not straighten; kept as delivered, nothing resampled. What is left is the worm's posture: the head swings and the brain bends with it, a neuron moving up to 86 px.
- Image. Channel 1, the neurons' nuclei and little else; channel 0, which also shows the pharynx, is left out, and one of the two changes with neural activity. Dim 8-bit counts that fade as the recording runs: a median of 14 at the points in the first volume, 5 in the last. No mask.
- Spacing. No voxel size is given, so distances are in pixels of the image plane. The step between planes was estimated at 3.6 pixels from the round bulbs of the pharynx; it may be a tenth off, more if the worm is pressed flat, and an error along z counts in the TRE through it.
- Truth. In 25 volumes all 178 neurons carry a person's tag; ten are the reference volumes the paper annotated in full by hand. A person's tag can mark a tracker result checked and accepted, so a method that works as ZephIR does may be favoured.
- Precision. The provider gives none. A point lies 0.6 px (median) from the intensity centre of its nucleus, with no systematic shift. In a volume of a test pair 0 to 10 of the 178 points lie within 2 px of another, two names on one nucleus, kept as delivered. One point of volume 672 lies half a plane outside the volume.
- Baseline. With no registration the 10 test pairs score a TRE of 15.5 px (4.1 to 34.8), its 95th percentile 26.3 px on average. On seven pairs most neurons lie farther from their place than their nearest neighbour (7 px). ZM9624_t0444_t0149, nearly one posture, leaves 4.1 px, still several times the points' 0.6 px.
- Pairs. Ours: the 25 volumes in the order of sha256("2026:ZephIRWorm:ZM9624:<volume>"), paired two by two, the first fixed, volume 298 left over and two pairs unused. Some volumes in different test pairs are seconds apart (95 and 100, 389 and 400, 444 and 463): other images of nearly one posture.
- Development. The second worm, a recording of Nguyen et al. as ZephIR's authors pass it on, has 79 tracks whose maker neither paper names: a truth to tune on, not graded. Its nuclei are in channel 0, its plane step is 4.5 pixels, and its pairs keep 13.5 px (9.5 to 22.4). Another strain and microscope: settings tuned there may not carry over.
- Ranked apart. The ten test pairs are twenty volumes of one worm in four minutes, the same 178 neurons in postures it takes again and again: ten samples of one worm's motion, not ten independent cases. The truth itself passes a main board's three checks.
- Outside numbers. None: ZephIR's paper measures the share of neurons tracked within their nucleus, with volumes of the recording annotated.
No method has a result on every pair of this dataset yet.
| Accuracy | Smoothness | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | TRE px | Change | rTRE | SDlogJ | Folds % | Time s | RAM GB | GPU GB | Images |
| No registration | 15.50 ±7.33 | 0.0213 ±0.0101 | 0.000 | 0.00 | ||||||
TRE: distance between matching landmarks. rTRE: the TRE over the length of the image's diagonal, to compare datasets of other sizes and units. SDlogJ: how irregular the deformation is, 0 is perfectly smooth. Folds: share of the image where the deformation folds over itself.
Pair by pair
Click a score to look at that result: the fixed image, and over it the moving image after registration, with the annotations. Underlined scores have their image; for the others you see the pair without the result. A green cell gained against no registration on that pair, a red one lost.
About the data
- Test pairs
- 10 test pairs drawn with seed 2026 among the 12 that the 25 fully annotated volumes of the ZM9624 worm make, each volume in one pair. The 10 development pairs come from the record's other worm, of Nguyen et al. 2017.
- Truth
- landmarks by experts the centres of the 178 neurons the recording tracks, each marked in both volumes of a pair: points a person placed in ZephIR's annotation interface, or results of the tracker that a person checked or corrected and took as their own
- Source
- zenodo.org/records/10818810
doi.org/10.5281/zenodo.10818810
doi.org/10.1371/journal.pcbi.1012075
github.com/venkatachalamlab/ZephIR
doi.org/10.1371/journal.pcbi.1005517 - Licence
- CC BY 4.0
10 pairs for tuning, never tested
Nguyen_t0114_t0961, Nguyen_t0189_t1343, Nguyen_t0475_t0640, Nguyen_t0659_t0553, Nguyen_t0964_t0335, Nguyen_t0968_t0284, Nguyen_t1184_t0106, Nguyen_t1227_t0338, Nguyen_t1298_t0816, Nguyen_t1368_t0742