Oral surgery, oral medicine, oral pathology and oral radiology · 2026
Robust classification of odontogenic cysts on panoramic radiographs using convolutional neural networks.
СтатьяНазвание 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.
Bouffleur Frederic, Westerdorf Julian, Scheurer Mats, Hoffmann Jürgen, Freudlsperger Christian, Eisenmann Urs, Ringwald Friedemann, Hagen Niclas, Kühle Reinald
Аннотация
OBJECTIVE: To develop an artificial intelligence for detecting and classifying odontogenic cysts on panoramic radiographs (PR).
STUDY DESIGN: A convolutional neural network (Inception-V3) was trained on 1424 PR, including 712 with histopathologically confirmed cysts (298 radicular cysts [RC], 231 dentigerous cysts [DC], 130 odontogenic keratocysts [OKC], 33 residual cysts, 20 ameloblastomas [AM], and 712 controls without cysts [NC]). Data augmentation and dropout were applied to reduce overfitting. During training, 10-fold cross-validation guided model and hyperparameter selection. Performance was assessed by accuracy, sensitivity, specificity, precision, F1 score, and confusion matrices. Gradient-weighted class activation maps were used to confirm cyst detection.
RESULTS: Binary classification (no cyst vs cyst) achieved an accuracy of 0.88 (sensitivity: 0.79, specificity: 0.97, precision: 0.97, F1-score: 0.87). Ternary classification (NC, RC, DC) reached 0.85 accuracy. Multi-class classification (NC, RC, DC, OKC/AM) yielded 0.67 accuracy, with reduced sensitivity for aggressive lesions.
CONCLUSIONS: Although using a single-center dataset with few PRs for rare pathologies, the model demonstrated strong performance in detecting and classifying odontogenic cysts, particularly for binary and benign differentiation. Furthermore, it proved robust to confound pathological findings. This high specificity within a heterogeneous control group shows promise as a supportive diagnostic tool in clinical practice.
STUDY DESIGN: A convolutional neural network (Inception-V3) was trained on 1424 PR, including 712 with histopathologically confirmed cysts (298 radicular cysts [RC], 231 dentigerous cysts [DC], 130 odontogenic keratocysts [OKC], 33 residual cysts, 20 ameloblastomas [AM], and 712 controls without cysts [NC]). Data augmentation and dropout were applied to reduce overfitting. During training, 10-fold cross-validation guided model and hyperparameter selection. Performance was assessed by accuracy, sensitivity, specificity, precision, F1 score, and confusion matrices. Gradient-weighted class activation maps were used to confirm cyst detection.
RESULTS: Binary classification (no cyst vs cyst) achieved an accuracy of 0.88 (sensitivity: 0.79, specificity: 0.97, precision: 0.97, F1-score: 0.87). Ternary classification (NC, RC, DC) reached 0.85 accuracy. Multi-class classification (NC, RC, DC, OKC/AM) yielded 0.67 accuracy, with reduced sensitivity for aggressive lesions.
CONCLUSIONS: Although using a single-center dataset with few PRs for rare pathologies, the model demonstrated strong performance in detecting and classifying odontogenic cysts, particularly for binary and benign differentiation. Furthermore, it proved robust to confound pathological findings. This high specificity within a heterogeneous control group shows promise as a supportive diagnostic tool in clinical practice.