An Interpretable Fuzzy-AI Clinical Decision Support System for Selecting Orthognathic Surgery in Skeletal Class III Malocclusion

This study aimed to develop an interpretable fuzzy-artificial intelligence (AI) framework to support treatment decision-making between orthognathic surgery and orthodontic camouflage in patients with skeletal class III malocclusion, while providing clinician-oriented explanations using a large language model (LLM). Eighty-six patients with skeletal class III malocclusion were classified into Surgery or Camouflage groups based on treatment plans determined by multiple experienced orthodontists. Pretreatment lateral cephalograms were analyzed, and 22 cephalometric variables were measured. A fuzzy inference model was used to calculate a membership grade (MG) representing the severity of skeletal class III sagittal discrepancy, based on visual evaluation of the facial profile. The MG was derived from a composite index incorporating 4 sagittal skeletal parameters: the ANB angle, chin-to-Nasion perpendicular distance, cranial base length, and mandibular body length. A decision-tree model was subsequently constructed using MG and selected cephalometric variables, and its discriminative performance was assessed using cross-validated area under the curve (AUC) analysis. To enhance interpretability, the mathematical decision framework was integrated with an LLM that generated concise, clinician-friendly diagnostic explanations without altering the underlying decision logic. Decision-tree analysis identified overjet, MG, lower facial height, and the H angle, an indicator of lip prominence, as the most influential predictors of treatment selection. The final model demonstrated excellent discriminative performance, with an AUC of 0.97. The LLM successfully translated the rule-based diagnostic output into brief explanatory summaries, and the complete system was implemented as open-source software. This LLM-integrated fuzzy-AI framework formalizes key elements of expert clinical reasoning and may improve the transparency, standardization, and reproducibility of treatment planning for skeletal class III malocclusion.