Abstract
Background Malocclusion is a prevalent dental disorder characterized by the improper alignment of the teeth and jaws, which adversely affects oral function and aesthetics. Traditional diagnosis and treatment planning procedures are time-consuming and rely heavily on expert evaluation. Recent developments in generative artificial intelligence (AI) offer promising tools to enhance diagnostic accuracy and to optimize treatment strategies in orthodontics. Methods This study developed a hybrid artificial AI model combining convolutional neural networks (CNNs), generative adversarial Networks (GANs), and large language models (LLMs), specifically, large language model Meta AI (LLaMA), to analyze orthodontic images and predict malocclusion types. A dataset comprising 150 anonymized, expert-labeled lateral cephalometric images from Hamad Dental Center in Doha was divided into training and testing sets using an 80:20 split. Data preprocessing, augmentation, and feature extraction were employed to improve model robustness. The model's performance was evaluated through classification accuracy and confidence scores. Results The hybrid model demonstrated high accuracy in predicting malocclusion classes, achieving confidence scores exceeding 85% across multiple test cases. Synthetic image augmentation via GAN improved the model's ability to generalize from limited data. The integration of LLM facilitated enhanced interpretation of clinical data, supporting precise treatment recommendations. Conclusion The generative AI-driven hybrid model effectively supports the diagnosis and treatment planning for malocclusion, thereby offering a valuable tool for orthodontic practice. Its ability to learn from limited data and provide high-confidence predictions streamlines clinical workflows. Future work will focus on expanding datasets, improving model explainability, and conducting clinical validation to ensure broader adoption in precision orthodontics.