Deep Learning-based tooth segmentation for enhanced visualization of dental anomalies and pathologies.
Authors
Affiliations (8)
Affiliations (8)
- Chongqing University Three Gorges Hospital, Chongqing University, Chongqing, China, 400044. Electronic address: [email protected].
- Department of Biotherapy, West China Hospital, Sichuan University, Chengdu, Sichuan, China, 610041. Electronic address: [email protected].
- State Key Laboratory of Oral Diseases & National Clinical Research Center for Oral Diseases, Department of Orthodontics, West China Hospital of Stomatology, Sichuan University, Chengdu, Sichuan, China, 610041. Electronic address: [email protected].
- Chengdu Boltzmann Intelligence Technology Co. Ltd, Chengdu, China, 610095. Electronic address: [email protected].
- Chengdu Boltzmann Intelligence Technology Co. Ltd, Chengdu, China, 610095. Electronic address: [email protected].
- State Key Laboratory of Oral Diseases & National Clinical Research Center for Oral Diseases, Department of Orthodontics, West China Hospital of Stomatology, Sichuan University, Chengdu, Sichuan, China, 610041. Electronic address: [email protected].
- State Key Laboratory of Oral Diseases & National Clinical Research Center for Oral Diseases, Department of Orthodontics, West China Hospital of Stomatology, Sichuan University, Chengdu, Sichuan, China, 610041. Electronic address: [email protected].
- State Key Laboratory of Oral Diseases & National Clinical Research Center for Oral Diseases, Department of Orthodontics, West China Hospital of Stomatology, Sichuan University, Chengdu, Sichuan, China, 610041. Electronic address: [email protected].
Abstract
This study aimed to developed and validated a deep-learning method for instance-level tooth segmentation in CBCT to enhance visualization and streamline detection of dental anomalies. The proposed deep learning model was trained in segmenting teeth in scans on data from 470 scans with various dental anomalies (e.g. caries, missing teeth, bone island, periapical periodontitis) or dental histories (e.g. filling, restoration, root canal surgery). Training involved an accelerated annotation procedure in which experts annotated some of the images in the dataset, which helped the model annotate the remaining images. Experienced dentists identified anomalies and pathologies in another 60 scans after manual interpretation or segmentation by the deep learning model. The trained model required 7.025 ± 2.885sec to segment teeth in a single scan with an accuracy of 0.934 ± 0.045 on the Jaccard index and mean relative volume difference of 0.075 ± 0.066. When aided by the segmentation overlays, dentists reduced anomaly-reading time by 20%. The proposed deep-learning framework achieves fully automated, instance-level segmentation of individual teeth in CBCT volumes with high geometric fidelity and clinically acceptable processing time. The high accuracy of the system supports its potential as a reliable tool in general dentistry.