Automatic deep learning detection of overhanging restorations in bitewing radiographs

Area of Science:

• Dentistry
• Artificial Intelligence
• Medical Imaging

Background:

• Overhanging dental restorations are a common issue in restorative dentistry.
• Accurate detection and segmentation are crucial for preventing secondary caries and periodontal disease.

Purpose of the Study:

• To assess the effectiveness of deep convolutional neural network (CNN) algorithms for detecting and segmenting overhanging dental restorations.
• To evaluate a You Only Look Once (YOLOv5) CNN model for this diagnostic task.

Main Methods:

• A dataset of 1160 anonymized bitewing radiographs was used.
• A CNN model (YOLOv5) was trained on 80% of the data, validated on 10%, and tested on 10%.
• Performance metrics including accuracy, sensitivity, precision, F1 score, and AUC were computed.

Main Results:

• The CNN model achieved a precision of 90.9%, sensitivity of 85.3%, and F1 score of 88.0%.
• The area under the receiver operating characteristic curve (AUC) was 0.859.
• The mean average precision (mAP) at an IoU threshold of 0.5 was 0.87.

Conclusions:

• Deep CNN algorithms demonstrate high effectiveness in detecting and diagnosing overhanging dental restorations.
• The study highlights the potential of deep learning in advancing dental diagnostic procedures.