Automated segmentation of dental CBCT image with prior-guided sequential random forests

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Craniomaxillofacial Surgery

Background:

• Cone-beam computed tomography (CBCT) is vital for diagnosing and planning treatments for craniomaxillofacial (CMF) deformities.
• Accurate 3D model generation from CBCT requires precise image segmentation.
• Image artifacts in CBCT present significant challenges for accurate segmentation.

Purpose of the Study:

• To develop and validate a novel, fully automated method for segmenting CBCT images.
• To overcome challenges posed by image artifacts in CBCT segmentation.
• To improve the accuracy of 3D model generation for CMF deformity diagnosis and treatment planning.

Main Methods:

• A majority voting method was used to estimate initial segmentation probability maps from expert-segmented CBCT images.
• Random forest classifiers were iteratively trained using CBCT appearance features and context features from probability maps.
• The method refines probability maps through successive classifier layers for enhanced segmentation accuracy.

Main Results:

• The automated segmentation method achieved high accuracy on CBCT scans from 30 subjects.
• Quantitative validation showed average Dice ratios of 0.94 for the mandible and 0.91 for the maxilla.
• Results significantly outperformed a state-of-the-art sparse representation method (p < 0.001).

Conclusions:

• A novel, fully automated method for CBCT segmentation has been successfully developed and validated.
• This method offers a significant advancement in the accurate segmentation of CBCT images for CMF applications.
• The improved accuracy facilitates better 3D model generation for clinical diagnosis and treatment planning.