A deep learning based automated maxillary sinus segmentation and bone grafts analysis in CBCT images

Xinrui Lang

Abstract: Precise evaluation of bone gain after maxillary sinus augmentation is critical for optimizing implant therapy yet manual measurements remain time consuming. This study validated a fully automated deep learning system named SA-ai to quantify bone augmentation. A paired CBCT dataset from 85 patients was used to train and test the system which integrates a 2D U-Net for sinus contour and a 3D V-Net for maxilla segmentation. The system achieved a Dice coefficient of 93.2% and registration RMSE of 1.046 mm. Clinical validation against manual measurements showed excellent agreement for bone volume (ICC = 0.993) and other parameters. Bias analysis confirmed measurement stability while workflow efficiency improved over 20-fold compared to manual methods. This registration subtraction paradigm delivers an automated and objective solution for longitudinal monitoring of bone graft volume including one-stage implant cases potentially standardizing clinical evaluation of post augmentation bone dynamics.