Automated Classification of Alveolar Bone Defects for Preoperative Augmentation Planning Using Deep Learning.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Imaging Analysis
• Oral and Maxillofacial Surgery

Background:

• Accurate assessment of alveolar bone deficiencies is crucial for successful dental implant surgery.
• Current methods for evaluating bone defects can be time-consuming and subjective.
• Advanced imaging like Cone Beam Computed Tomography (CBCT) provides detailed anatomical information.

Purpose of the Study:

• To develop and evaluate a deep learning framework for automated detection and classification of alveolar ridge deficiencies.
• To compare the diagnostic performance of four different Convolutional Neural Network (CNN) architectures.
• To assess the clinical utility of an AI-driven system in preoperative dental implant planning.

Main Methods:

• A dataset of 1305 CBCT cross-sectional images was curated and labeled into four categories: healthy, horizontal defect, vertical defect, and combined defect.
• Four CNN models (RegNetY-008, EfficientNetV2-S, ResNet50, MobileNetV3-Large) were trained and evaluated.
• Performance metrics included accuracy, weighted precision, recall, F1-score, and epoch duration.

Main Results:

• The RegNetY-008 model achieved the highest accuracy (93.87%) and weighted F1-score (93.88%), with the fastest processing time (8.04 sec/epoch).
• EfficientNetV2-S followed with 93.10% accuracy.
• RegNetY-008 demonstrated superior ability in classifying complex combined defects with minimal errors.

Conclusions:

• Deep learning models, particularly RegNetY, can effectively classify alveolar bone defects from CBCT images.
• The automated system offers a rapid, objective tool for clinicians, enhancing preoperative planning for dental implants.
• This technology can assist in treatment decisions, potentially reducing complications and planning time.