PRECISION BASED CRANIOFACIAL SURGERY: AI ENABLED MOLECULAR PREDICTION OF CRANIOSYNOSTOSIS RELAPSE

Abstract

Background: Postoperative relapse represents a persistent challenge in the surgical treatment of craniosynostosis and is influenced by a complex interplay of clinical, radiographic, and molecular determinants. Reliable preoperative risk assessment is therefore essential to improve surgical decision-making and long-term outcomes. This study aimed to develop and validate an integrated artificial intelligence (AI)–based model that utilizes multimodal data to predict postoperative relapse in patients with craniosynostosis.

Methods: A retrospective–prospective translational study was performed involving 120 patients with syndromic and nonsyndromic craniosynostosis who underwent primary craniofacial surgery and completed a minimum follow-up period of 24 months. Comprehensive clinical information, radiographic features, and molecular genetic data targeting established craniosynostosis-related genes were collected. Supervised machine learning algorithms were employed to construct predictive models with stepwise integration of clinical, radiographic, and molecular variables. Model performance was evaluated using accuracy, sensitivity, and specificity, positive and negative predictive values, F1 score, and area under the receiver operating characteristic curve (ROC) (AUC).

Results: Postoperative relapse was observed in 27 patients (22.5%), whereas 93 patients (77.5%) remained relapse-free. A significant association was identified between the presence of pathogenic genetic variants and relapse occurrence (p = 0.008). The fully integrated clinical, radiographic, molecular AI model demonstrated superior predictive performance, achieving an accuracy of 92.5%, sensitivity of 88.9%, specificity of 94.7%, F1 score of 87.8%, and an AUC of 0.946. The model correctly identified 88.9% of patients who developed relapse and accurately classified 94.6% of patients with favorable postoperative outcomes. Progressive improvements in predictive accuracy were noted with the incremental incorporation of multimodal data.

Conclusion: The integrated AI-based predictive framework exhibited excellent accuracy and discriminative capability for postoperative relapse prediction in craniosynostosis. The inclusion of molecular genetic information alongside clinical and radiographic parameters substantially enhanced risk stratification, supporting the potential role of AI-driven models as clinical decision-support tools for personalized surgical planning and precision-based management of craniosynostosis.