Scan Body Optimisation for Enhancing Full‐Arch Implant Scanning Accuracy Using Smartphone Video and a Deep Learning Model: An In Vitro Study

OBJECTIVE: A multi-view 3D reconstruction deep learning (DL) model was used to convert smartphone videos of a full-arch implant cast into 3D scans. This study aimed to determine whether a custom scan body (SB) would outperform regular stock SB in this DL scenario, compared with conventional impressions and photogrammetry. MATERIALS AND METHODS: A maxillary edentulous model with six implant analogues was scanned using four protocols (n = 10/group): conventional splinted open-tray impression (CO), photogrammetry (PG; Icam4D), DL using stock PEEK SBs (DLS), and DL using custom SBs (DLC). DL scans were reconstructed from smartphone videos using Neuralangelo. The custom SBs were 3D-printed with geometric features and spray-coated. Trueness and precision were assessed by root mean square (RMS), linear, and angular deviations. RESULTS: DLC outperformed DLS in trueness (RMS: 51.1 ± 7.6 vs. 80.6 ± 27.8 μm; linear: 78.0 ± 43.5 vs. 144.1 ± 83.4 μm) and precision (RMS: 50.5 ± 9.6 vs. 75.3 ± 19.7 μm; linear: 78.9 ± 43.5 vs. 122.8 ± 71.3 μm; all p < 0.001). CO and PG exhibited the highest RMS (23.1 ± 5.4 vs. 27.1 ± 1.3 μm; p = 0.93) and linear (32.4 ± 14.1 vs. 35.7 ± 31.4 μm; p = 0.663) trueness. PG achieved the best precision across RMS (7.9 ± 3.3 μm), linear (8.7 ± 4.2 μm) and angular (0.07° ± 0.05°) measurements. CONCLUSION: Custom SB significantly improves the accuracy of DL-based full-arch implant scanning compared with PEEK stock SB. While early stage, this DL approach shows potential as a viable option for full-arch implant scanning.