Microcomputed tomography analysis of marginal adaptation and internal fit of zirconia frameworks with direct-to-abutment connections using a deep learning artificial intelligence model

STATEMENT OF PROBLEM: Despite the increased use of direct-to-abutment connections for implant-supported fixed complete dental prostheses (ISFCDPs), limited evidence exists regarding marginal adaptation and internal fit, 2 parameters critical for biomechanical performance. PURPOSE: This study evaluated the marginal adaptation and internal fit of zirconia frameworks with direct-to-abutment connections compared to conventional titanium base (Ti-base) connections. A novel method for internal fit assessment using a deep learning artificial intelligence (AI) model was also introduced, and level of agreement with traditional measurements was assessed. Internal fit was compared across the 3 designs using both traditional and AI-derived measurements. MATERIAL AND METHODS: Zirconia specimens (n=24) were allocated to 3 groups: frameworks with Ti-base and regular screws (n=8), frameworks fabricated direct-to-abutment with regular screws (n=8), and frameworks fabricated direct-to-abutment with modified screws (n=8). Each specimen underwent microcomputed tomography scanning before and after thermomechanical aging (TMA) to assess internal fit and marginal adaptation. A deep learning-based AI segmentation model was developed using an image processing software program and achieved a final validation loss below 0.01, corresponding to a Dice Similarity Coefficient greater than 0.95. This metric indicates high spatial overlap and robust segmentation of the gap and abutment regions. Volumetric measurements were obtained for 1 straight and 1 angled transmucosal abutment per specimen. Specimens were subjected to 1.2 million loading cycles at 90 N, followed by repeated measurements. Ninety-six sites were evaluated before and after TMA. Two measurements each for straight and angled abutments for the 3 designs were averaged and difference scores were calculated as post-TMA minus pre-TMA values. Nonparametric statistical tests were applied because of heterogeneity of group variances. Post hoc Mann-Whitney U tests, intraclass correlation and Bland-Altman analyses were conducted (α=.05). RESULTS: Significant differences in marginal adaptation among the 3 groups were identified before and after TMA. The Ti-base group demonstrated significantly superior marginal adaptation compared to both direct-to-abutment groups at both time points (z=3.36, P<.001). Significant differences in internal fit were also observed among the 3 groups before and after TMA. Ti-base group showed significantly improved internal fit relative to both direct-to-abutment groups (standardized Mann-Whitney z values ranged from 3.15 to 3.36, P<.001). Intraclass correlations revealed a significant level of agreement between standardized measurements obtained using conventional and AI-based methods (P<.001), and Bland-Altman plots provided support for concurrent validity of the AI-derived measures of internal fit. CONCLUSIONS: Ti-base connections in ISFCDPs demonstrated superior marginal adaptation and internal fit compared to direct-to-abutment connections. Both connection designs achieved clinically acceptable internal fit values. For marginal adaptation, the direct-to-abutment regular screw group exceeded the clinically acceptable threshold of 120 µm. A significant level of agreement was found between standardized traditional and AI-derived measurements of internal fit. Bland-Altman plots supported the concurrent validity of the deep learning AI model.