Affiliations: [a] School of Odonto Stomatology, Hanoi Medical University, Hanoi, Vietnam | [b] Faculty of Computer Science and Engineering, Thuyloi University, Hanoi, Vietnam | [c] School of Information and Communication Technology, Hanoi University of Science and Technology, Hanoi, Vietnam | [d] VNU Information Technology Institute, Vietnam National University, Hanoi, Vietnam
Correspondence:
[*]
Corresponding author. Le Hoang Son, VNU Information Technology Institute, Vietnam National University, Hanoi 010000, Vietnam. E-mail: [email protected].
Abstract: Periapical Inflammation (PI) is one of the most popular diseases in adults due to complication of endodontitis or dental trauma with corresponding consequences to quality-of-life like tiredness and signs of infection. Specifically, patients having severe PI are often tiredness and high fever accompanied by signs of infection such as dry lips, dirty tongue, lymph node reaction in the area under the jaw. In X-Ray images, PI could be recognized by vague boundaries with signs of periapical ligament extensions. It is necessary to design a computerized diagnosis system based on the Deep Learning models for supporting clinicians in diagnosis of PI from X-Ray images. In this paper, we propose a new medical system called VNU for diagnosis of PI from X-Rays images. The VNU system uses Deep Learning to classify whether X-Ray images being PI or not. The Residual Neural Network (ResNet) with 34 layers is utilized with proper data augmentation and learning algorithms. The system is designed based on 7-layer enterprise architecture (User, Business, Application, Data, Technology, Infrastructure, and Security). It is used by both the clinicians and IT operators. The system has been validated on real data from Hanoi Medical University, Vietnam consisting of 900 images with PI and 500 normal images. Two scenarios of validation namely hyperparameter selection and performance comparison with other CNN-based Deep Learning models have been performed. It has been found from the experiments that the proposed system has better performance than the others in terms of sensitivity and specificity with the corresponding values of 96.70% and 93.87%. The system is deployed on the web interface that offers flexibility for clinicians in diagnosis and training.
Keywords: Apical lesions, periapical radiograph, ResNet, deep learning, VNU system
