Can dental fillings affect the performance of an AI-driven innovative tool for automatic tooth segmentation in cone-beam computed tomography: A validation study?

Abstract

To assess the performance of an innovative artificial intelligence (AI)-driven tool for segmentation of teeth with filling materials on cone-beam computed tomography (CBCT). After training of an AI network with 175 CBCT images, 74 CBCT images were divided into two groups: an experimental group (50 CBCT images - with at least 1 tooth with fillings) and a control group (24 CBCT images – with all sound teeth). Each group was composed of 113 teeth paired according to the type of tooth: 40 molars, 40 premolars, and 33 anterior teeth. The AI-driven tool based on a convolutional neural network automatically detected and segmented the teeth. Two experienced dental surgeons reviewed segmentation quality and performed refinements if needed. The main outcome was segmentation consistency. Also, time analysis was performed from the random selection of 10% of the teeth that needed refinements within each type of teeth according to the method of segmentation (AI, AI+refinements, and manual). The AI-tool showed excellent range of values regarding the consistency of the segmentation for the control (Hausdorff Distance (HD)-0.02-0.25; Intersection over union (IoU)-0.97-0.99; Precision-1.00; Recall-0.97-0.99; Accuracy-1.00) and the experimental groups (HD:0.32-0.48; IoU:0.91-0.95; Precision:1.00; Recall:0.91-0.95; Accuracy:0.99-1.00) regardless of the type of tooth. Tooth segmentation based on the AI-driven tool had the lowest working time with a mean time of 29.8 seconds ( p <0.05), which was 36.6 times faster than manual segmentation. The proposed AI-driven tool showed the highly accurate and fast performance to carry out the teeth segmentation regardless of the presence of dental filling material.