RESUMEN

OBJECTIVES: To evaluate the performance of smartphone scanning applications (apps) in acquiring 3D meshes of cleft palate models. Secondarily, to validate a machine learning (ML) tool for computing automated presurgical plate (PSP). MATERIALS AND METHODS: We conducted a comparative analysis of two apps on 15 cleft palate models: five unilateral cleft lip and palate (UCLP), five bilateral cleft lip and palate (BCLP) and five isolated cleft palate (ICP). The scans were performed with and without a mirror to simulate intraoral acquisition. The 3D reconstructions were compared to control reconstructions acquired using a professional intraoral scanner using open-source software. RESULTS: Thirty 3D scans were acquired by each app, totalling 60 scans. The main findings were in the UCLP sample, where the KIRI scans without a mirror (0.22 ± 0.03 mm) had a good performance with a deviation from the ground truth comparable to the control group (0.14 ± 0.13 mm) (p = .653). Scaniverse scans with a mirror showed the lowest accuracy of all the samples. The ML tool was able to predict the landmarks and automatically generate the plates, except in ICP models. KIRI scans' plates showed better performance with (0.22 ± 0.06 mm) and without mirror (0.18 ± 0.05 mm), being comparable with controls (0.16 ± 0.08 mm) (p = .954 and p = .439, respectively). CONCLUSIONS: KIRI Engine performed better in scanning UCLP models without a mirror. The ML tool showed a high capability for morphology recognition and automated PSP generation.