Embryo Stage Classification Data with Expert and AI Model Comparisons

245 single-frame embryo images were classified by three human embryologists and two deep learning models, ResNet-34 and VGG16. The study, authored by Radhika Kakulavarapu and last updated in March 2026, evaluates accuracy, agreement, and interpretability using explainable AI techniques like Grad-CAM. Embryologists achieved 89.9% accuracy, outperforming the AI models, while interpretability assessments showed ResNet-34 explanations were rated biologically relevant 89% of the time.

Use Cases

• Benchmarking AI model accuracy against human expert performance based on embryo stage classification results.
• Analyzing inter-operator agreement between embryologists and deep learning models using kappa statistics.
• Evaluating the biological relevance of model-generated explanations like Grad-CAMs for interpretability.
• Investigating the relationship between model accuracy and spatial attention overlap in explanations.

Strengths

• Includes direct comparisons of accuracy between three human embryologists (89.9%) and two deep learning models (74.3% and 78.8%).
• Provides quantitative agreement metrics (κ≥0.932) and stage-wise agreement analysis.
• Assesses interpretability with specific percentages for biologically relevant explanations (89% for ResNet-34 vs. 59% for VGG16).

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data is from a single-center, retrospective study, which may limit generalizability.

Provenance

Source

Radhika Kakulavarapu via figshare.

Collection Method

Retrospective study using single-frame embryo images classified by embryologists and deep learning models.

Time Range

null

Freshness

Last updated 2026-03-18 10:00:40; freshness should be verified.

Geography

null