Performance evaluation of deep learning models for the classification and identification of dental implants

Area of Science:

• Artificial Intelligence in Dentistry
• Deep Learning for Medical Imaging
• Object Detection Models

Background:

• Deep learning for dental implant classification is established.
• Object detection models for implant design identification lack comprehensive accuracy investigations.

Purpose of the Study:

• To evaluate an object detection deep learning model's performance in classifying 103 dental implant designs.
• To assess the model's accuracy across different implant thirds (coronal, middle, apical).

Main Methods:

• Extracted 14,037 implant images from panoramic radiographs.
• Utilized You Only Look Once (YOLO) versions 5 and 7 algorithms for model training and evaluation.
• Applied data augmentation via image processing and generative adversarial networks to enhance accuracy.

Main Results:

• YOLOv7 achieved higher mean average precision (mAP) than YOLOv5 across three cross-validation datasets.
• Image processing and generative adversarial networks significantly improved mAP, reaching up to 0.988.
• The model demonstrated high accuracy in classifying 26 implant design classes.

Conclusions:

• Object detection models show high accuracy for classifying dental implant designs.
• Model performance is influenced by algorithm choice, image processing techniques, and implant design complexity.