A benchmark dataset for deformable registration and object matching of biological images
Introduced in Rotation-invariant graph transformers enable global bioimage registration and cell matching
| Sample name | Modality | N nuclei | Volume shape (pixels) |
|---|---|---|---|
| stmhc_pl1 | fluorescence, confocal | 10108 | 768x2054x1310 |
| stmhc_pl2 | fluorescence, confocal | 9496 | 768x2054x1310 |
| stmhc_pl3 | fluorescence, confocal | 10252 | 748x2070x1440 |
| stmhc_pl4 | fluorescence, confocal | 9857 | 729x1980x1330 |
| stmhc_pl5 | fluorescence, confocal | 10219 | 706x2050x1370 |
| em | SBEM | 11488 | 1224x1962x1785 |
match_gt.csv: table listing corresponding nuclei between samplesmatch_gt.csv: table listing corresponding nuclei between samples, left and right nuclei exchanged{sample name}.csv- label: segmentation mask ID
- z, y, x: coordinates measured in pixels
- mean_stmhc: mean intensity of StMhc staining within the cell mask. For EM calculated within nuclei masks instead of cells
- reg_gt: annotated nuclei ID from
match_gt.csv, otherwise-1(same ID for the same nucleus in different samples) - reg_gt_reflected: annotated nuclei ID from
match_gt.csv, otherwise-1
{sample name}.ome.zarr- / : CZYX volume with two channels: DAPI and StMhc staining. For EM: raw EM volume and muscle segmentation
- labels
- nuclei_segm
- cell_segm
- reg_gt:
Point clouds and matching ground truth annotations can be found in data. Images can be downloaded from BioImage Archive:
For data viewing and metrics install platy5p1
conda env create -f environment.yml
conda activate platy5p1_env
pip install -e .
For examples of working with the dataset see
-
examples/matching_evaluation.ipynb -
examples/view_images.ipynb
examples/matching contains baseline matching used in the publication.