DentalSegmentator: Robust open source deep learning-based CT and CBCT image segmentation.

Area of Science:

• Digital dentistry
• Medical imaging analysis
• 3D reconstruction

Background:

• Accurate segmentation of anatomical structures in dento-maxillo-facial (DMF) computed tomography (CT) and cone beam computed tomography (CBCT) scans is crucial for digital dentistry.
• Existing methods may lack automation or robustness, hindering widespread clinical adoption.

Purpose of the Study:

• To introduce and evaluate DentalSegmentator, a novel open-source tool for fully automatic segmentation of five critical anatomical structures on DMF CT and CBCT scans.
• To assess the performance and generalizability of DentalSegmentator across diverse datasets.

Main Methods:

• A retrospective dataset of 470 CT and CBCT scans was used for training and validation.
• The tool's performance was evaluated on an internal dataset (133 scans) and an external dataset (123 scans) by comparing automatic segmentations with expert segmentations.
• Key performance metrics included Dice Similarity Coefficient (DSC) and Normalized Surface Distance (NSD).

Main Results:

• High accuracy was achieved, with mean overall DSC of 92.2% on the internal dataset and 94.2% on the external dataset.
• Mean overall NSD was 98.2% on the internal dataset and 98.4% on the external dataset, indicating precise surface distance.
• The results demonstrate robust multiclass segmentation capabilities across a diverse range of DMF CT and CBCT scans.

Conclusions:

• DentalSegmentator provides a fully automatic, robust, and accurate solution for segmenting anatomical structures in DMF CT and CBCT scans.
• The open-source tool, available as a 3D Slicer extension, facilitates the creation of patient-specific 3D models for various digital dentistry applications.
• This approach supports visualization, treatment planning, and intervention in clinical practice.