PANDIA: Cross-Dataset Infant Pain Assessment Model Results

87.3% accuracy was achieved by the PANDIA AI system for infant pain assessment across four datasets. The results, published by Oussama El Othmani in May 2026, evaluate a multimodal system combining hierarchical learning, graph reasoning, meta-learning, and symbolic explanations on data from 2,847 infants.

Use Cases

• Benchmarking multimodal AI models for clinical tasks based on the described hierarchical representation learning and graph-based fusion.
• Evaluating model generalization across heterogeneous medical datasets based on the four-dataset evaluation.
• Studying federated learning frameworks for privacy-preserving multi-site collaboration as described.
• Analyzing the trade-off between model accuracy and interpretability based on the concept-bottleneck and symbolic reasoning approach.
• Testing meta-learning adaptation for personalized assessment with minimal per-subject data as outlined.

Strengths

• Model achieved 87.3% accuracy and a 92.1% clinician acceptance rate for explanations.
• Evaluation involved 2,847 infants across four datasets.
• The system maintained fewer than 30M parameters, suitable for edge deployment.
• All code, trained models, and preprocessing pipelines are publicly available.

Limitations

• Dataset heterogeneity across collection sites is noted as a limitation.
• The validation design is retrospective, requiring prospective clinical trials before live deployment.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Oussama El Othmani

Collection Method

Results from evaluating the PANDIA AI system on four infant pain assessment datasets.

Freshness

Last updated 2026-05-26 17:32:03.