Reconocimiento y medición automáticos de estructuras anatómicas asociadas con la elevación del piso del seno maxilar...

The purpose of this study is to develop a deep learning model that can identify the maxillary sinus, posterior superior alveolar artery(PSAA), and alveolar ridge, and evaluate its diagnostic performance. Based on this, relevant parameters for preoperative design of maxillary sinus elevation can be measured to achieve intelligent preoperative design for maxillary posterior tooth implantation surgery.

A total of 2400 CBCT slices from patients with maxillary posterior tooth loss was selected as the initial dataset. Anatomical structure annotation and enhanced YOLOv11 architecture were used for model training to achieve segmentation of maxillary sinus, PSAA, and alveolar ridge. Intersection over union (IoU), average precision (AP), average recall (AR) and the Euclidean distance were used to evaluate the accuracy of structure segmentation. On the basis of the segmentation of the three important anatomical structures mentioned above, five anatomical parameters (A1-A5) related to maxillary posterior tooth implantation were set, and their errors were statistically analyzed.

The median IoU for maxillary sinus segmentation was 0.945 (IQR: 0.934-0.951, 95%CI: 0.935-0.941), while the median IoU for PSAA segmentation was 0.991 (IQR: 0.982-1.000, 95%CI: 0.948-0.974). The model achieved an average precision of 0.902 ± 0.023 and a recall of 0.937 ± 0.024 for PSAA segmentation. For alveolar crest localization, the mean Euclidean distance errors between predicted and ground-truth landmarks were 0.50 ± 0.31 mm and 0.38 ± 0.24 mm for the two key points, respectively. 95% of AI prediction errors for A1-A4 were within 1 mm, while 95% of AI prediction errors for A5 were within 10 mm².

The enhanced YOLOv11 framework reliably and autonomously identifies critical anatomical structures for maxillary sinus elevation including the maxillary sinus, PSAA, and maxillary alveolar crest in CBCT images. This model enables the acquisition of reliable clinical parameters, demonstrating its potential for future intelligent assisted preoperative evaluation and design of maxillary posterior dental implant surgery.