A Convolutional Neural Network Approach for Dental Panoramic Radiographs Classification

Abstract

Dental Panoramic Radiograph (DPR) has been widely accepted in the dental industry in recent years for providing valuable information to dental practitioners. Conventional fashion of analyzing this information has always been relying on the professional individuals. Nonetheless, the recent technology breakthrough with computer vision is promising in reducing effort and labor costs. The purpose of this study is to propose an automated DPR classification method that could be used to identify condition such as having undergone endodontic therapy. This new method used a combination of unique preprocessing procedure with highly developed image recognition algorithm known as Convolutional Neural Network (CNN). The new method firstly preprocesses and normalizes the DPR images as the training dataset, then trains the network model using the training data. Finally, testing datasets will then be used for generating evaluation results. Based from the experimental results, the trained network with preprocessed training datasets performed an accuracy rate of 85.3%. In addition, the performance of the network could still be improved by increasing the number of training datasets and using more preprocessing methods. In conclusion, computer visions can provide confident better performance in DPR classification in practices such as DPR classification.