Assessment of artificial intelligence–based remote monitoring of clear aligner therapy: A prospective study

Introduction: Artificial intelligence remote monitoring of clear aligner therapy has recently gained popularity. It uses deep learning algorithms on a patient's mobile smartphone to determine readiness to progress to the next aligner (ie, GO vs NO-GO) and identify areas in which the teeth are not tracking with the clear aligners. This study aimed to assess the repeatability of the Go or No-Go instructions provided by the application and to determine the 3-dimensional discrepancies that constitute an unseat. Methods: Thirty patients in treatment with clear aligners at an academic clinic were scanned twice using a remote monitoring application on a smartphone, and the results were compared. Gauge repeatability and reproducibility analysis were performed. Intraoral and remote monitoring scans were obtained on the same day from 24 additional clear aligner patients that completed treatment using their final aligners. The intraoral scan after using the final aligner and the stereolithography file of the planned position at the final aligner was compared with measure the maximum discrepancies between the actual and planned position of the teeth. Results: Gauge compatibility of 44.7% was noted. In total 83.3% of patient instructions agreed between Scan 1 and 2, but 0% agreed completely on which and/or how many teeth had tracking issues. Patients who received GO instruction had mean greatest discrepancies of 1.997 mm, 1.901 mm, 0.530 mm, 8.911°, 7.827°, and 7.049° in mesiodistal, buccolingual, occlusogingival, tip, torque, and rotational dimensions, respectively. These discrepancies were not significantly different from patients receiving NO-GO instruction (1.771 mm, 1.808 mm, 0.606 mm, 8.673°, 8.134°, and 6.719° for the corresponding categories). Conclusions: Despite the study's limitations, these findings suggest concerns with the consistency of remote monitoring instructions because of gauge compatibility over the industry standard. Similarly, large discrepancies in tooth position for patients receiving GO and NO-GO instruction suggest that artificial intelligence decisions were inconsistent with quantitative findings.