ZFScanner斑马鱼发育筛选模型与结果表
数据集简介
Description
This deposit provides trained model weights and analysis result tables for the ZFScanner zebrafish developmental-screening pipeline. Together with the well-image dataset (10.5281/zenodo.21868073) and the analysis code (https://github.com/...), it enables reproduction of all figures in the associated paper except for the representative time-lapse photography.
Contents
The dataset is organized into the following directories:
models/
Contains five trained neural network weights:
- Two U-Net segmentation models (for embryo segmentation)
- One ten-point landmark regressor (for morphological landmark detection)
- Two ResNet-18 screening models (for developmental stage classification)
csv/
Contains analysis result tables, with qinfo.csv as the main table summarizing quantitative measurements across all embryos and conditions.
manual-labels/
Contains manual annotations including segmentation masks and landmark coordinates used for training and validation.
blind-predictions/
Contains predictions from the blind sensitivity analysis, used to assess model robustness.
lda-results/
Contains per-condition screening tables derived from linear discriminant analysis (LDA) for developmental stage classification.
Reproducing the figures
To reproduce the figures from the paper:
- Download the well-image dataset (10.5281/zenodo.21868073).
- Download this deposit (models, tables, manual labels).
- Run the analysis code from the GitHub repository (https://github.com/...).
The provided model weights and tables are sufficient to regenerate all quantitative results and plots, except for the representative time-lapse images.
Files
The deposit includes the following files:
README.md: Description and usage instructions.checksums.md5: MD5 checksums for data integrity verification.zfscanner-manual-labels.tar(226 MB): Manual annotation files.zfscanner-models.tar(192 MB): Trained model weights.zfscanner-tables.tar.gz(76 MB): Analysis result tables (csv/, lda-results/, blind-predictions/).
Additional details
- Total size: 491 MB (compressed), ~0.7 GB (unpacked).
- License: Data is licensed under CC BY 4.0; code is under MIT License.
- Publication date: August 12, 2026.
- Repository: Zenodo (open access, free download).
Related works
- Well-image dataset: 10.5281/zenodo.21868073
- Analysis code: https://github.com/...
- Original paper: "ZFScanner: high-throughput digital phenotyping of zebrafish development using an AI-driven office scanner"
Keywords and subjects
- Zebrafish (Danio rerio)
- Developmental screening
- Deep learning
- Phenotyping
- Flatbed scanner
- U-Net, ResNet-18