Fig 1.
Cropped panoramic radiography of the individual lesion.
(A) Stafne’s bone cavity, (B) Dentigerous cyst, (C) Odontogenic keratocyst, (D) Ameloblastoma.
Table 1.
Number of image data used in this study and the percentage of images according to individual panoramic radiograph.
Fig 2.
Schematic representation of the pre-processing steps in data preparation.
Fig 3.
Characteristics of Dense block in DenseNet.
Table 2.
Structure and characteristics of DenseNet121 based convolutional neural network classifier for panoramic radiography.
Fig 4.
Confusion matrix of Stafne’s bone cavities classification from cysts and tumors using test data set.
(A) true value (B) normalized value. TP: true positive, FP: False positive, FN: false negative, TN: true negative.
Table 3.
Performance comparison of various models.
Fig 5.
The panoramic image and the importance-weighted visualization image of classification criteria (Grad-Cam and Guided Grad-Cam) in Stafne’s bone cavity (A), dentigerous cyst (B), odontogenic keratocyst, (C), and ameloblastoma (D). Note that the important degree of imaging features for Stafne’s bone cavity classification is color-coded from red (highly-weighted) to blue (less-weighted). The model visualizes the empty internal area and mandibular inferior cortex in Stafne’s bone cavity, while tooth-bearing, multiple locules, and root resorption are well recognized in cysts and tumors.