Artificial Intelligence for Classifying the Relationship between Impacted Third Molar and Mandibular Canal on...

Area of Science:

• Dental Radiology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Assessing the relationship between the lower third molar (MM3) and the mandibular canal (MC) is crucial for surgical planning.
• Panoramic radiography is a common imaging modality, but interpretation accuracy can vary, especially among inexperienced observers.
• Deep convolutional neural networks (CNNs) offer potential for automated image analysis in dentistry.

Purpose of the Study:

• To evaluate the diagnostic performance of ResNet-152 and VGG-19 CNNs in analyzing the MM3-MC relationship on panoramic images.
• To compare the diagnostic capabilities of these CNNs against an inexperienced dental observer.

Main Methods:

• 142 MM3 images from 83 panoramic radiographs were used, with 80% for training/validation and 20% for testing.
• K-fold cross-validation was employed, and images were labeled by an experienced radiologist.
• Diagnostic accuracy, sensitivity, specificity, and positive predictive value (PPV) were calculated for CNNs and a dental student.

Main Results:

• ResNet-152 achieved high performance: 84.09% sensitivity, 94.11% specificity, 92.11% PPV, and 88.86% accuracy.
• VGG-19 demonstrated good results: 71.82% sensitivity, 93.33% specificity, 92.26% PPV, and 85.28% accuracy.
• The dental student's performance was significantly lower, with 69.60% sensitivity, 53.00% specificity, 64.85% PPV, and 62.53% accuracy.

Conclusions:

• Deep CNN architectures like ResNet-152 show significant potential for identifying and evaluating the MM3-MC contact in panoramic images.
• CNNs can serve as valuable tools to enhance the diagnostic accuracy of inexperienced observers in interpreting panoramic dental radiographs.
• AI-assisted analysis may improve the reliability of MM3-MC relationship assessment, aiding in safer dental procedures.