Deploying a novel deep learning framework for segmentation of specific anatomical structures on cone-beam CT

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Cone-beam computed tomography (CBCT) is vital in dentistry for diagnostics and treatment planning.
• Automatic prediction of anatomical structures on CBCT images can improve efficiency and accuracy.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for automatic anatomical structure segmentation on CBCT images.
• To assess the model's performance in delineating key dental and maxillofacial structures.

Main Methods:

• A deep learning model (nnUNetv2) was trained on 63 CBCT datasets (405 slices each).
• Anatomical structures were annotated by dentomaxillofacial radiologists.
• The model was evaluated using accuracy, Dice score, precision, and recall on 7 test datasets.

Main Results:

• The model achieved high accuracy (0.99) for multiple structures, including the nasal fossa and maxillary sinus.
• Dice scores varied, with the maxillary sinus (0.98) performing best and the mandibular canal (0.73) performing lowest.
• Precision ranged from 0.73 to 0.98, indicating reliable segmentation.

Conclusions:

• The deep learning model demonstrates robust performance in segmenting anatomical features on CBCT images.
• The findings suggest significant potential for improving dental diagnostics and treatment planning.
• Further refinement may enhance segmentation consistency across all structures.