Dataset DOI: 10.5061/dryad.fbg79cp9z

Description of the data and file structure

This dataset contains raw data used to evaluate (1) the accuracy and reliability of embryonic developmental stage classification and (2) the identification of non-viable eggs using observational features. The data include observer-based classifications, ground-truth labels, and diagnostic features recorded. The observations are based on candling images taken of eggs from the captive crested ibis population during the 2025 breeding season at the Crested Ibis Breeding Center, Dongzhai National Nature Reserve, Henan Province, China

Files and variables

File: stage_validation.csv

Description: This file contains data used to validate the classification of embryonic developmental stages

Variables

• date: Date of candling
• egg_id: Identity of egg
• day: Incubation process (day)
• old_name: The original filename of image, including information of candling date and incubation day
• image: New filename that generated to conceal candling date and incubation day
• true_stage: True developmental stage (ordinal variable)
• observer_1: Stage assigned by observer 1
• observer_2: Stage assigned by observer 2
• observer_3: Stage assigned by observer 3

File: fertile_embryo_46.csv

Description: This dataset contains data used to validate the identification of fertilization status during Day 4-6 of incubation, along with associated diagnostic features

Variables

• image: The filename of image
• egg: Identity of egg
• type: The category that based on the final incubation outcomes and the developmental conditions (i.e., infertile, dead embryo, successfully hatched)
• day: Incubation process (day)
• fertility_true: True fertility status (0 = infertile, 1 = fertile)
• fer_obs1: Fertility classification by observer 1
• fer_obs2: Fertility classification by observer 2
• fer_obs3: Fertility classification by observer 3
• s_broad_shadow1: A broad shadow when candled from the sharp end. Observation by observer 1 (0 = absence, 1 = presence)
• s_shadow_with_distinct_edge1: Distinct shadow edge when candled from the sharp end. Observation by observer 1 (0 = absence, 1 = presence)
• b_broad_shadow1: A broad shadow when candled from the blunt end. Observation by observer 1 (0 = absence, 1 = presence)
• b_shadow_with_distinct_edge1: Distinct shadow edge when candled from the blunt end. Observation by observer 1 (0 = absence, 1 = presence)
• s_broad_shadow2: A broad shadow when candled from the sharp end. Observation by observer 2 (0 = absence, 1 = presence)
• s_shadow_with_distinct_edge2: Distinct shadow edge when candled from the sharp end. Observation by observer 2 (0 = absence, 1 = presence)
• b_broad_shadow2: A broad shadow when candled from the blunt end. Observation by observer 2 (0 = absence, 1 = presence)
• b_shadow_with_distinct_edge2: Distinct shadow edge when candled from the blunt end. Observation by observer 2 (0 = absence, 1 = presence)
• s_broad_shadow3: A broad shadow when candled from the sharp end. Observation by observer 3 (0 = absence, 1 = presence)
• s_shadow_with_distinct_edge3: Distinct shadow edge when candled from the sharp end. Observation by observer 3 (0 = absence, 1 = presence)
• b_broad_shadow3: A broad shadow when candled from the blunt end. Observation by observer 3 (0 = absence, 1 = presence)
• b_shadow_with_distinct_edge3: Distinct shadow edge when candled from the blunt end. Observation by observer 3 (0 = absence, 1 = presence)

File: fertile_embryo_o6.csv

Description: This dataset contains data used to validate the identification of fertilization status and embryo viability during Day 6-24 of incubation, along with associated diagnostic features

Variables

• image: The filename of image
• egg: Identity of egg
• type: The category that based on the final incubation outcomes and the developmental conditions (i.e., infertile, dead embryo, successfully hatched)
• day: Incubation process (day)
• fertility_true: True fertility status (0 = infertile, 1 = fertile)
• fer_obs1: Fertility classification by observer 1
• fer_obs2: Fertility classification by observer 2
• fer_obs3: Fertility classification by observer 3
• die_day: Time of death for dead embryo
• embryo_true: True embryo viability (0 = dead, 1 = alive; NA for unfertilized eggs)
• em_obs1: Embryo viability classification by observer 1
• em_obs2: Embryo viability classification by observer 2
• em_obs3: Embryo viability classification by observer 3
• s_broad_shadow1: A broad shadow when candled from the sharp end. Observation by observer 1 (0 = absence, 1 = presence)
• s_shadow_with_distinct_edge1: Distinct shadow edge when candled from the sharp end. Observation by observer 1 (0 = absence, 1 = presence)
• s_shadow_coverage_gt501: Shadow coverage exceeding 50% when candled from the sharp end. Observation by observer 1 (0 = absence, 1 = presence)
• s_uniform_dark_shadow1: A uniform dark shadow when candled from the sharp end. Observation by observer 1 (0 = absence, 1 = presence)
• s_vascular_structures1: Vascular structures when candled from the sharp end. Observation by observer 1 (0 = absence, 1 = presence)
• b_broad_shadow1: A broad shadow when candled from the blunt end. Observation by observer 1 (0 = absence, 1 = presence)
• b_shadow_with_distinct_edge1: Distinct shadow edge when candled from the blunt end. Observation by observer 1 (0 = absence, 1 = presence)
• b_vascular_structures1: Vascular structures when candled from the blunt end. Observation by observer 1 (0 = absence, 1 = presence)
• b_aircell_with_distinct_edge1: An air cell with distinct edge. Observation by observer 1 (0 = absence, 1 = presence)
• s_broad_shadow2: A broad shadow when candled from the sharp end. Observation by observer 2 (0 = absence, 1 = presence)
• s_shadow_with_distinct_edge2: Distinct shadow edge when candled from the sharp end. Observation by observer 2 (0 = absence, 1 = presence)
• s_shadow_coverage_gt502: Shadow coverage exceeding 50% when candled from the sharp end. Observation by observer 2 (0 = absence, 1 = presence)
• s_uniform_dark_shadow2: A uniform dark shadow when candled from the sharp end. Observation by observer 2 (0 = absence, 1 = presence)
• s_vascular_structures2: Vascular structures when candled from the sharp end. Observation by observer 2 (0 = absence, 1 = presence)
• b_broad_shadow2: A broad shadow when candled from the blunt end. Observation by observer 2 (0 = absence, 1 = presence)
• b_shadow_with_distinct_edge2: Distinct shadow edge when candled from the blunt end. Observation by observer 2 (0 = absence, 1 = presence)
• b_vascular_structures2: Vascular structures when candled from the blunt end. Observation by observer 2 (0 = absence, 1 = presence)
• b_aircell_with_distinct_edge2: An air cell with distinct edge. Observation by observer 2 (0 = absence, 1 = presence)
• s_broad_shadow3: A broad shadow when candled from the sharp end. Observation by observer 3 (0 = absence, 1 = presence)
• s_shadow_with_distinct_edge3: Distinct shadow edge when candled from the sharp end. Observation by observer 3 (0 = absence, 1 = presence)
• s_shadow_coverage_gt503: Shadow coverage exceeding 50% when candled from the sharp end. Observation by observer 3 (0 = absence, 1 = presence)
• s_uniform_dark_shadow3: A uniform dark shadow when candled from the sharp end. Observation by observer 3 (0 = absence, 1 = presence)
• s_vascular_structures3: Vascular structures when candled from the sharp end. Observation by observer 3 (0 = absence, 1 = presence)
• b_broad_shadow3: A broad shadow when candled from the blunt end. Observation by observer 3 (0 = absence, 1 = presence)
• b_shadow_with_distinct_edge3: Distinct shadow edge when candled from the blunt end. Observation by observer 3 (0 = absence, 1 = presence)
• b_vascular_structures3: Vascular structures when candled from the blunt end. Observation by observer 3 (0 = absence, 1 = presence)
• b_aircell_with_distinct_edge3: An air cell with distinct edge. Observation by observer 3 (0 = absence, 1 = presence)

File: Validation_of_features.R

Description: The provided R script performs the following analyses:

1. Classification of Developmental Stage

• Calculates classification accuracy per observer and overall mean accuracy
• Computes mean absolute error between observed and true stages
• Assesses agreement using:
• Weighted Cohen’s kappa (observer vs. true stage)
• Krippendorff’s α (multi-observer agreement)
• Pairwise weighted Cohen’s kappa between observers

2. Identification of Non-viable Eggs

• Computes classification accuracy for fertility and embryo viability
• Summarizes mean accuracy across observers

• Calculates:
• Sensitivity
• Specificity
• Precision
• Computes confusion matrix components (TP, TN, FP, FN)
• Summarizes performance across observers

• Fleiss’ kappa for multi-observer agreement
• Pairwise Cohen’s kappa between observers

3. Validation of diagnostic features

• Uses generalized linear mixed models (GLMM; binomial family)
• Models relationships between diagnostic features and:
• Fertility status
• Embryo viability
• Includes observer as a random effect

Notes

• Missing values ("NA") indicate unavailable observations.
• For embryo viability analyses, only records with non-missing "embryo_true" were included.
• Some analyses require at least two or three observers per egg (e.g., Fleiss’ kappa).

Code/software

All analyses were conducted in R 4.5.1 using the following packages: irr, tidyr, car, dplyr, stringr, and lme4