Validation of an AI Model for Automated Detection of Alveolar Bone Changes Post-orthodontics Using Cone-Beam Computed Tomography

利用锥形束计算机断层扫描技术验证用于自动检测正畸后牙槽骨变化的AI模型

阅读:1

Abstract

INTRODUCTION: With recent advances in artificial intelligence (AI), image-based tools are increasingly being explored to assist clinicians in detecting changes in alveolar bone levels following orthodontic treatment. This retrospective, cross-sectional study aimed to develop and validate an automated AI model for quantifying alveolar bone levels using cone-beam computed tomography (CBCT) images from patients who underwent fixed orthodontic therapy, with the objective of curating a robust dataset, training AI models for landmark detection, evaluating accuracy against manual annotations, and assessing clinical applicability. MATERIALS AND METHODS: At the Department of Orthodontics, 1,200 CBCT images from 200 patients (104 (52%) males, 96 (48%) females; mean age 23.6 ± 2.37 years) yielded 3,600 interdental sites. Images covering the maxillary and mandibular segments were preprocessed using median filtering and contrast-limited adaptive histogram equalization. The AI pipeline employed You Only Look Once, version 8 (YOLOv8; Ultralytics, Frederick, MD, USA) for tooth segmentation (720 training, 240 validation, and 240 test images) and a mask region-based convolutional neural network (Mask R-CNN) for generating tooth, bone, and crown masks. Custom algorithms were developed to localize the cementoenamel junction (CEJ) and alveolar bone crest (ALC). A random forest regression model was then applied to identify predictors of alveolar bone loss. RESULTS: The cohort of 200 patients (104 males, 52%; 96 females, 48%; mean age 23.6 ± 2.37 years) showed alveolar bone loss in 124 patients (62%). The mean periodontal index was 1.90 ± 0.57 mm, and the mean orthodontic treatment duration was 26.72 ± 4.41 months. The mean ALC levels were 1.35 ± 0.65 mm in the anterior region and 1.47 ± 0.62 mm in the posterior region. YOLOv8 achieved a mean average precision (mAP) of 0.941 at an intersection-over-union (IoU) threshold of 0.5 (mAP50) and a tooth detection accuracy of 95.44% at complete IoU. Post-augmentation, Mask R-CNN yielded accuracies of 92.28% (tooth mask), 94.15% (bone mask), and 94.11% (crown mask). CEJ precision ranged from 85.3% to 94.5%, and ALC precision ranged from 82.8% to 95.3%, with a root mean square error of 0.023-0.071. Random forest regression identified treatment duration (importance score: 0.321) as the primary predictor of bone loss. CONCLUSION: This AI-driven pipeline offers an efficient and accurate tool for periodontal monitoring, enabling early detection of bone changes and supporting personalized post-orthodontic care. Future prospective studies with diverse cohorts are required to enhance clinical integration.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。