Trueness of artificial intelligence-driven CBCT tooth segmentation: A comparative validation ex vivo pilot study

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate segmentation of dental structures from cone-beam computed tomography (CBCT) is crucial for diagnosis and treatment planning.
• AI-driven tools offer potential for automating and improving segmentation accuracy.
• Comparative validation against physical ground truth is essential for assessing AI performance.

Purpose of the Study:

• To evaluate and compare the trueness of four commercial AI segmentation tools (Diagnocat, Relu, CephX, CoDiagnostiX) and one open-source software (3D Slicer with DentalSegmentator).
• To assess the accuracy of these tools for dental CBCT segmentation using extracted human teeth as a physical ground truth reference.

Main Methods:

• Ten single-rooted human teeth were scanned using high-resolution CBCT.
• Extracted teeth were digitized using an intraoral scanner to create ground-truth reference models.
• DICOM datasets were processed by each software, and 3D models were registered and compared using surface-to-surface distance metrics (e.g., Hausdorff, RMSD, MAD).

Main Results:

• Significant differences in segmentation trueness were observed across all tools (p<0.001).
• Relu (MAD 0.10±0.02 mm) and Diagnocat (MAD 0.10±0.04 mm) demonstrated the highest trueness.
• CephX showed the lowest trueness (MAD 0.25±0.05 mm), while 3D Slicer performed comparably to commercial solutions.

Conclusions:

• Relu and Diagnocat achieved the highest trueness in CBCT tooth segmentation against physical specimens.
• All tested tools exhibited clinically acceptable trueness for most applications, but performance varied significantly.
• Further large-scale validation studies are needed to investigate performance differences before definitive clinical recommendations.