Accuracy and time efficiency of deep learning-based method for single-crown design compared with a conventional CAD...

Designing restoration contours is a labor-intensive and time-consuming process. Although computer-aided design and computer-aided manufacturing (CAD-CAM) systems have demonstrated potential for automating occlusal morphology in single-tooth crowns, their accuracy compared with deep learning (DL)-based approaches remains unclear.

This systematic review aimed to investigate the accuracy and time-efficiency of DL-based software programs in reproducing single-tooth crown designs compared with those from conventional CAD systems without DL support.

This study followed the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) statement. Electronic searches were carried out on 5 databases up to July 2025. Only in vitro studies comparing DL-based software program and conventional CAD systems in terms of single-tooth crown morphology reconstruction and worktime were included. Two examiners independently evaluated the methodological quality of the studies using the methodological index for nonrandomized studies (MINORS) scale.

Seven studies met the eligibility criteria for inclusion. The accuracy of restoration contour design and finish line detection by DL-equipped software program was better than or comparable with that of conventional CAD software program without DL support, regardless of operator intervention or experience. Internal fit exhibited discrepancies within clinically acceptable values, ranging from 55.4 µm to 83.1 µm (P<.001). Working time for the single-tooth crown design was significantly lower for the DL-based software program (P<.05).

The DL-based software program achieved accuracy higher than or comparable with the conventional CAD software program without DL assistance in terms of occlusal morphology and finish line design, while also requiring a shorter working time.