Automated cone cut error detection of bitewing images using convolutional neural network

Introduction: Cone cut error is one of the technical errors that can hinder the important information from a bitewing radiograph. Meanwhile, deep learning is a specialized artificial intelligence method where an algorithm can be trained to automatically detect, classify and give output based on the trained dataset. Aims: This research aimed to apply deep learning methods to classify cone cut errors from bitewing radiographs. Methods: This study received ethical approval from the IIUM Research Ethics committee: Approval no: IREC 2022-151. 2712 bitewing images were collected and classified into normal and error groups from the Imaging Unit, Kulliyyah of Dentistry, IIUM. The deep learning method selected was Convolutional Neural Network (CNN), and the algorithm was used and trained to classify the cone cut error. Data augmentation was used to increase the amount of data for training, validation, and testing. Results: The test dataset showed good results of 0.93-1.00 recall and precision, while scored 0.96 for the F1 score. Several modifications were made to tailor overfit and unbalanced data groups to get optimum results. Conclusion: This research experimented with an automated approach to utilize deep learning as a method of quality assessment in the dental radiology field, especially in bitewing images. The rapid advance of artificial intelligence should be utilized to improve the imaging quality of a dental radiograph