Convolutional neural network for automatic maxillary sinus segmentation on cone-beam computed tomographic images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate 3D segmentation of the maxillary sinus is vital for dental diagnostics and treatment planning.
• Manual segmentation of maxillary sinuses from cone-beam computed tomography (CBCT) datasets is time-consuming and labor-intensive.
• Convolutional neural networks (CNNs) have demonstrated significant potential in 3D medical image analysis.

Purpose of the Study:

• To develop and validate a novel automated CNN-based methodology for maxillary sinus segmentation using CBCT images.
• To compare the automated method with semi-automatic segmentation regarding time efficiency, accuracy, and consistency.
• To assess the reliability of the automated segmentation through inter-observer analysis.

Main Methods:

• A dataset of 264 maxillary sinuses from CBCT scans was utilized, divided into training, validation, and testing sets.
• A 3D U-Net architecture CNN model was developed for automated segmentation.
• The automated segmentation was compared against semi-automatic methods and evaluated using Dice Similarity Coefficient (DSC).

Main Results:

• Automated segmentation reduced processing time from 60.8 minutes to 0.4 minutes (p < 2.2e-16).
• The CNN model achieved a high segmentation accuracy with a DSC of 98.4%.
• Inter-observer reliability for minor manual refinements showed an excellent DSC of 99.6%.

Conclusions:

• The proposed CNN model offers a time-efficient, precise, and consistent automated solution for maxillary sinus segmentation.
• This methodology facilitates accurate 3D model generation for improved diagnosis and virtual treatment planning.
• The automated approach has the potential to streamline clinical workflows in dentistry and radiology.