Automatic Detection of Dentigerous Cysts on Panoramic Radiographs: A Deep Learning Study

Aim: The aim of this study is to create a model that enables the detection dentigerous cysts on panoramic radiographs in order enable dentistry students meet and apply artificial intelligence applications.
 Methods: E.O. I.T. who are 5th year faculty dentistry, detected 36 orthopantomographs whose histopathological examinations were determined as Dentigerous Cyst, affected teeth cystic cavities segmented using CranioCatch's supported clinical decision support system software. Since sizes images dataset different from each other, all resized 1024x514 augmented vertical flip, horizontal flip both flips applied train-validation. Within obtained data set, 200 epochs trained with PyTorch U-Net learning rate 0.001, train: 112 (112 labels), val: 16 (16 labels). With created after segmentations completed, new cyst tested success was evaluated.
 Results: for cysts, F1 score (2TP / (2TP+FP+FN)) precision (TP/ (TP+N)) sensitivity (TP+FN)) found be 0.67, 0.5 1, respectively.
 Conclusion: CNN approach analysis images, has been even small database. These methods can improved, graduate dentists gain experience save time diagnosis lesions radiographs.