Accuracy of artificial intelligence-based functions in intraoral scanners: A systematic review.

STATEMENT OF PROBLEM: Artificial intelligence (AI)-based applications have increasingly been developed and integrated in different digital data acquisition technologies, including intraoral scanners (IOSs). A comprehensive evaluation of their accuracy and current state is lacking. PURPOSE: The purpose of this systematic review was to evaluate the accuracy of AI-based applications integrated within IOSs to optimize digital data acquisition. MATERIAL AND METHODS: A literature search was completed in 4 databases: PubMed/Medline, Embase, Web of Science, and Cochrane. A manual search was also conducted. Studies evaluating the accuracy of AI-based applications in IOSs to optimize the scanning process were included. Two investigators evaluated the studies independently by applying the Joanna Briggs Institute critical appraisal. A third examiner was consulted to resolve any lack of consensus. The included articles were organized based on the AI-based applications tested: AI cleaning, automatic alignment with implant scan body (ISB) computer-aided design (CAD) file, and occlusal collisions correction of articulated scans in maximum intercuspation position (MIP). RESULTS: Seven articles were included. One study evaluated an AI-assisted scanning application and reported an overall improvement in scanning accuracy, evidenced by a lower root mean square (RMS) error in complete arch scans. Two studies assessed AI-based ISB file registration and reported conflicting results, with 1 study demonstrating improved accuracy and the other reporting reduced accuracy. However, differences in ISB design and implant scanning techniques were present among these studies. Finally, 4 studies evaluated occlusal collision correction; 3 reported improved MIP accuracy in completely dentate patients, whereas the remaining study reported variable outcomes depending on the extent of partial edentulism. CONCLUSIONS: The available scientific evidence supporting these AI‑based applications remains limited. The performance of these AI‑based applications appears to be limited to the specific IOS system and generation evaluated and demonstrated variable performance depending on their intended function.