PANDIA: Infant Pain Assessment Performance Across Four Datasets

A dataset containing out-of-distribution performance analysis for a multimodal AI system designed for infant pain assessment. The system, PANDIA, was evaluated on 2,847 infants across four datasets and achieved 87.3% accuracy. The dataset was created by Oussama El Othmani and last updated on May 26, 2026.

Use Cases

• Benchmarking multimodal AI models for infant pain assessment based on the described accuracy metrics.
• Evaluating model robustness on out-of-distribution test sets based on the performance analysis.
• Studying federated learning frameworks for privacy-preserving multi-site clinical collaboration.
• Analyzing the trade-off between model size and accuracy for edge deployment in clinical settings.
• Investigating clinician acceptance of AI explanations based on the reported 92.1% acceptance rate.

Strengths

• Performance metrics are explicitly reported, including 87.3% accuracy and a 12.4% improvement over the best baseline.
• The dataset is associated with a fully publicly available code repository, trained models, and preprocessing pipelines.
• The model was evaluated on a substantial cohort of 2,847 infants across four datasets.

Limitations

• The specific data columns and row count are unknown, limiting detailed suitability assessment.
• The dataset is described as retrospective, with noted heterogeneity across collection sites.
• The primary clinical dataset (NICU-MM) is available only upon request subject to an ethical data use agreement.

Provenance

Source

figshare

Collection Method

Likely contains performance metrics and analysis from the evaluation of the PANDIA AI system.

Freshness

Last updated 2026-05-26 17:32:14