Computerized Analysis of Panoramic Images for Automatic Detection and Classification of Dental Reconstructions
1Applied Physics, Jerusalem College of Technology, Israel
2Oral Medicine, Maxillofacial Imaging,, Hadassah University Hospital, Israel
3Radiology, Hadassah University Hospital, Israel
PURPOSE: A panoramic radiograph is an external scanning dental radiograph of the whole oral cavity, obtained with minimal discomfort and significantly lower radiation dose compared to intra-oral radiographs. However, it requires careful review by a dentist with the proper expertise in order to diagnose pathologies. The aim of this study was to develop a prototype of a comprehensive computer tool designed to automatically map the oro-maxillofacial structures in a panoramic radiograph. This will allow segmentation and classification of dental reconstructions (fillings, crowns, root canal treatments, implants), in order to aid the clinician in patient`s management
MATERIAL AND METHODS: An algorithm, based on computer vision and machine learning, was developed to automatically segment and classify dental reconstructions in a panoramic radiograph. Panoramic anonymized images were automatically cropped to obtain a region of interest, containing the upper and lower jaws. Initially, the algorithm used a gradient image to highlight the borders of the various structures. The structures were then automatically segmented using a local adaptive threshold to obtain a binary image in which the required structures were highlighted in white. In order to improve detection of these structures, computer imaging tools such as erosion, dilation, opening and closing were used. Since each structure was characterized by a unique shape and gray level distribution, 20 numerical features describing contour and texture were extracted to classify the structures. The training of the algorithm was performed on 14 panoramic images that included 253 dental structures. These structures were classified into 8 different types, including crowns, root canals, fillings, implants, etc. 22 different machine learning models for automatic classification were evaluated by a Cross-Validation process.
RESULTS: Several machine learning models yielded good classification of the dental reconstructions. Three types of Support Vector Machine (SVM) models yielded the most successful classification results. The “Cubic SVM” model resulted in an accuracy of 97%, the "Quadratic SVM" model resulted in an accuracy of 95%, and "Medium Gaussian SVM in an accuracy of 91%.
CONCLUSION: Algorithm was used for automatic detection and classification of dental reconstructions in panoramic imaging. This algorithm may be extrapolated for future study to detect bone pathologies in panoramic images using greater number of cases in the training dataset.