A novel AI-driven tool for automated root canal segmentation of single and bi-rooted teeth on cone-beam computed tomography

Purpose To develop and validate a novel artificial intelligence (AI)-driven tool for automated root canal (RC) segmentation in single and bi-rooted teeth on cone-beam computed tomography (CBCT). Methods A total of 81 CBCT scans acquired from two devices with distinct protocols were collected and randomly split into the training (n=65; 183 teeth) and validation (n=16; 32 teeth) of the AI networks. Afterwards, 61 CBCT scans (120 single and 70 bi-rooted teeth) were employed to test the performance of the developed AI-driven tool. The CBCT scans from the testing sample were automatically segmented, and the resulting three-dimensional (3D) RC models were exported in standard triangle language format. An experienced oral and maxillofacial radiologist assessed the quality of the segmentation and made refinements to create refined-AI 3D models (AI-R). The performance of the AI tool was assessed by comparing the AI and AI-R models. Additionally, 30% of the testing sample was randomly chosen to assess the time consumed for performing three different segmentation methods (manual, AI and AI-R). Results The AI-driven tool exhibited highly accurate RC segmentation for single teeth (Dice similarity coefficient (DSC): 89-93%; 95% Hausdorff distance (HD): 0.10-0.13 mm) and bi-rooted teeth (DSC: 88-93%, 95%HD: 0.13-0.16 mm). In terms of time analysis, AI segmentation proved to be the fastest method, taking 42±10.5 s (p<0.05), marking a 64-fold reduction compared to manual segmentation (2687±815.7 s). Conclusions The novel AI-driven tool showed a highly accurate and fast performance for segmenting the root canal of single and bi-rooted teeth on CBCT scans.