Prediction of Permanent Tooth-crowding Using Panoramic-radiographs of the Mixed Dentition

Methods: A total of 300 panoramic radiograph images recorded from 100 patients, who had not experienced active orthodontic treatment but were under observation, were selected. For each patient, records of one or more mixed and one permanent dentition stages were included. Images of the lower teeth on the right and left sides were processed separately, providing 600 feature vectors. The vectors were categorized into two groups according to the visual examination of the panoramic radiographs of the lower permanent dentition: “well-aligned with minimum tooth crowding or spacing” and “malaligned designated by the overlapping of tooth crowns greater than a 1/4 of the crown width or the discrepancy between proximal tooth contacts greater than a 1/4 of the crown height.” A coordinate system adaptive to each image was defined based on positions and sizes of lower incisor and molar teeth. Twenty-one dimensional feature vectors relevant to the eruption of lower permanent canine and premolar teeth, such as location, orientation and size of the teeth were generated. Template matching technique was applied to predict whether there is tooth crowding in the permanent dentition on the basis of information obtained from the mixed dentition records. Randomly selected 40 feature vectors in the mixed dentition category were used as the test data, with the remaining 560 as the training data. The trials were repeated 1000 times.

Results: The mean prediction accuracy for 1000 trials was 83%.

Conclusions: The proposed system for predicting permanent tooth crowding condition from panoramic radiographs of the mixed dentition was found to be reliable for possible clinical use.