Evaluation of deep learning MRI reconstruction for dental implant crowns in a phantom study

Abstract

Deep learning (DL) reconstruction is increasingly applied in clinical magnetic resonance imaging (MRI) to improve image quality and reduce scan time, but its impact on dental metal artifacts remains unclear. This pilot phantom study evaluated DL reconstruction compared with conventional reconstruction for various implant crowns. Acrylic phantoms containing titanium implants with four crown types-zirconia, PMMA, gold, and Ni-Cr metal-were scanned on a 3.0-T MRI system. Axial T1- and T2-weighted sequences were acquired using identical imaging parameters. Image quality (noise and signal-to-noise ratio [SNR]) and metal artifacts (visual scores and artifact ratio) were evaluated in the slice showing the largest crown area. DL reconstruction consistently reduced noise and improved SNR across all crown types and sequences. Metal artifact severity followed the material-dependent order: zirconia < PMMA < gold < Ni-Cr metal, in both sequences. Visual assessment showed no difference in artifact severity between DL and conventional images. DL reduced artifacts only in zirconia crowns on T2-weighted sequence (10.38% vs. 9.31%). These findings indicate that although DL reconstruction enhances overall image quality, its effectiveness in reducing dental metal artifacts remains limited. As this is a pilot study using phantoms, further in vivo validation is necessary.