Evaluating The Impact Of Voxel Size On AI-Based Segmentation Accuracy For Pulp Chamber And Root Canal Detection: An In Vitro Study

Objective: To evaluate the influence of cone-beam computed tomography (CBCT) voxel size on the accuracy of artificial intelligence (AI)-based segmentation of pulp chambers and root canals, using high-resolution CBCT as the reference standard, and to assess the effect of tooth type on segmentation performance. Methods: An in vitro study design was used to assess AI-based segmentation across three voxel sizes (75 µm, 150 µm, and 300 µm) and multiple tooth types, including incisors, premolars, and molars. Segmentation outputs were compared with direct reference-standard measurements from high-resolution CBCT scans. The influence of voxel size and anatomical complexity on performance was analyzed. Statistical analyses were performed to evaluate the effects of voxel size and tooth type on segmentation accuracy. Results: AI-based segmentation demonstrated high reliability and good agreement with reference CBCT measurements for canal length, with a mean absolute difference of approximately 0.5 mm. Segmentation performance was significantly influenced by both voxel size and tooth type. Reduced accuracy and consistent underestimation of canal number were observed in molars across all voxel sizes, highlighting challenges in anatomically complex structures. Segmentation completeness was also influenced by voxel size and tooth type; however, inter-observer agreement for this measure was low. Conclusions: AI-based segmentation demonstrates promising potential for accurate and efficient analysis of root canal anatomy in CBCT imaging. However, performance is influenced by imaging parameters and anatomical complexity. An intermediate voxel size (150 µm) may provide an optimal balance between image quality, radiation dose, and segmentation reliability. Further refinement of AI algorithms is needed to improve performance in complex anatomical regions and support clinical integration. Clinical Significance: AI-assisted segmentation may enhance diagnostic consistency and efficiency in endodontic imaging while reducing reliance on ultra–high-resolution CBCT protocols.

Author dc:creator

• Corso, Anne M.

Advisors dc:contributor.advisor

• Dutra, Vinicius
• Ehrlich, Ygal

• Segmentation
• Aritifical Intelligence
• Root Canal Detection

Language dc:language.iso

en_US

Handle dc:identifier.uri

https://hdl.handle.net/1805/56531

OAI identifier oai:identifier

oai:scholarworks.indianapolis.iu.edu:1805/56531