Using Deep Learning and Natural Language Processing Models to Detect Child Physical Abuse

This study developed deep learning models to identify child physical abuse using clinical data from 1,312 pediatric trauma patients. A model incorporating natural language processing of radiology reports achieved 93.4% accuracy in distinguishing abuse from accidental trauma, suggesting potential for real-time clinical decision support.