An Automated Framework for Mandibular Reconstruction: Evaluation and Clinical Application.

Abstract

Mandibular reconstruction suffers from limitations in automation and objectivity. This study aimed to develop an automated framework to address these challenges. A novel approach combining statistical shape modeling and 3D deep learning was developed. Trained on 200 CT scans and validated in 80 clinical cases, its performance was compared to expert manual planning by evaluating reconstruction accuracy, bone contact area, and planning time. Finally, design experiments were performed in five clinical cases. The automated method demonstrated high accuracy (DSC: 0.874) and real-time efficiency. It significantly outperformed manual planning in bone contact area (106.2 vs. 94.6 mm²) and reduced planning time by 85%. Postoperative results showed excellent implant alignment (2.1 mm deviation) and high patient satisfaction. The proposed automated framework successfully enhances the efficiency and quality of mandibular reconstruction, meeting clinical needs. Future work will integrate more complex biomechanical considerations.