Multi-task reinforcement learning and explainable AI-Driven platform for personalized planning and clinical decision support in orthodontic-orthognathic treatment

Abstract

Abstract This study presents a novel clinical decision support platform for orthodontic-orthognathic treatment that integrates multi-task reinforcement learning with explainable artificial intelligence. The platform addresses the challenges of personalized treatment planning in complex dentofacial deformities by formulating treatment as a sequential decision-making process optimizing multiple clinical objectives simultaneously. We developed a comprehensive framework comprising: (1) a multi-task reinforcement learning core with specialized state-action representations for craniofacial structures; (2) complementary explainable AI components that render complex model decisions interpretable within clinical contexts; and (3) an interactive interface facilitating collaborative human-AI decision-making. Experimental validation with 347 retrospectively analyzed cases demonstrated significant improvements in treatment plan quality (19.9%), decision efficiency (73.9% time reduction), and prediction accuracy (92.7%) compared to conventional methods. Clinical evaluation by multidisciplinary specialists confirmed the system’s practical utility, with particularly strong performance in complex cases featuring multiple dentofacial abnormalities. The proposed platform represents a significant advancement toward evidence-driven, personalized treatment planning in orthodontic-orthognathic therapy while maintaining clinical interpretability and expert oversight.