Evaluation of Cone-Beam Computed Tomography Images with Artificial Intelligence

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Cone-Beam Computed Tomography (CBCT)

Background:

• Interpreting CBCT images is complex, time-consuming, and requires specialized expertise.
• Digital transformation in healthcare necessitates advanced tools for efficient image analysis.
• AI-driven decision support systems can potentially reduce physician workload.

Purpose of the Study:

• To evaluate the performance of AI models based on convolutional neural networks (CNNs) for analyzing CBCT images.
• To assess the accuracy of AI in identifying 15 distinct dental and jaw conditions.
• To explore the utility of AI as a decision support tool in radiological interpretation.

Main Methods:

• Developed AI models using the Mask R-CNN architecture and ResNet 101 with transfer learning.
• Utilized a dataset of 300 CBCT images, segmented to label 15 conditions (e.g., caries, implants, lesions).
• Calculated model performance using the confusion matrix and F1 scores.

Main Results:

• The AI models demonstrated high success rates across all 15 evaluated conditions.
• The highest F1 score (0.99) was achieved for identifying dental implants.
• The lowest F1 score (0.8) was observed for mixed-appearing jaw lesions, with other conditions scoring between 0.87 and 0.98.

Conclusions:

• CNN-based AI models show significant potential for accurate analysis of CBCT images.
• These AI systems can effectively identify a wide range of dental pathologies and conditions.
• AI-powered image evaluation can serve as a valuable decision support tool, improving efficiency in radiological workflows.