Journal of dentistry · 2026
Artificial Intelligence assisted diagnosis of impacted teeth other than third molars: A systematic review.
СтатьяСистематический обзорНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Yu Lei, Zhao Zuodong, Jacobs Reinhilde, EzEldeen Mostafa, Cadenas de Llano Pérula Maria
Аннотация
OBJECTIVE: To systematically evaluate current Artificial Intelligence (AI) based approaches for the diagnosis of impacted teeth other than third molars, and to assess their diagnostic performance, clinical relevance, and existing limitations.
METHODS: PubMed, Web of Science, Cochrane Library, Embase, and Scopus were used to identify relevant studies. Study methodology, dataset preparation, and key metrics were collected from each included article. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) was applied.
RESULTS: Of the 30 included studies, 15 focused on mesiodens and 15 on impacted canines. No AI-based diagnostic studies on impacted premolars or incisors were identified. Panoramic radiographs were the most commonly used imaging modality, followed by CBCT and periapical radiographs. Across detection and classification tasks, most AI models demonstrated good to excellent diagnostic performance, with more than 70% of reported metrics exceeding 0.80. YOLO- and ResNet-based architectures were most frequently applied, with YOLO models generally achieving comparatively higher accuracy. Segmentation models showed high spatial agreement, with Dice similarity coefficients frequently exceeding 0.90. CBCT-based studies demonstrated strong performance for three-dimensional assessment, particularly for segmentation tasks, while alternative diagnostic materials and ANN-based approaches yielded moderate to high predictive accuracy for unerupted tooth size.
CONCLUSIONS: AI-based diagnostic systems show promising potential in the detection, classification, segmentation and prediction of impacted teeth other than third molars, offering valuable support for early intervention and personalized treatment planning. Future research should aim to include larger datasets, potentially through multi-center collaboration, as well as following standardized evaluation protocols, using more easily interpretable AI techniques (such as Grad-CAM, SHAP) to ensure robust, transparent, and clinically reliable implementation.
CLINICAL SIGNIFICANCE: AI provides accurate and efficient tools for detecting and classifying impacted teeth, enabling early diagnosis and personalized treatment planning. Its integration may improve diagnostic Accuracy and streamline orthodontic workflows, but broader adoption requires multi-center validation and explainable models to ensure reliability.
METHODS: PubMed, Web of Science, Cochrane Library, Embase, and Scopus were used to identify relevant studies. Study methodology, dataset preparation, and key metrics were collected from each included article. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) was applied.
RESULTS: Of the 30 included studies, 15 focused on mesiodens and 15 on impacted canines. No AI-based diagnostic studies on impacted premolars or incisors were identified. Panoramic radiographs were the most commonly used imaging modality, followed by CBCT and periapical radiographs. Across detection and classification tasks, most AI models demonstrated good to excellent diagnostic performance, with more than 70% of reported metrics exceeding 0.80. YOLO- and ResNet-based architectures were most frequently applied, with YOLO models generally achieving comparatively higher accuracy. Segmentation models showed high spatial agreement, with Dice similarity coefficients frequently exceeding 0.90. CBCT-based studies demonstrated strong performance for three-dimensional assessment, particularly for segmentation tasks, while alternative diagnostic materials and ANN-based approaches yielded moderate to high predictive accuracy for unerupted tooth size.
CONCLUSIONS: AI-based diagnostic systems show promising potential in the detection, classification, segmentation and prediction of impacted teeth other than third molars, offering valuable support for early intervention and personalized treatment planning. Future research should aim to include larger datasets, potentially through multi-center collaboration, as well as following standardized evaluation protocols, using more easily interpretable AI techniques (such as Grad-CAM, SHAP) to ensure robust, transparent, and clinically reliable implementation.
CLINICAL SIGNIFICANCE: AI provides accurate and efficient tools for detecting and classifying impacted teeth, enabling early diagnosis and personalized treatment planning. Its integration may improve diagnostic Accuracy and streamline orthodontic workflows, but broader adoption requires multi-center validation and explainable models to ensure reliability.