Automated Detection System Based on Convolution Neural Networks for Retained Root, Endodontic Treated Teeth, and Implant Recognition on Dental Panoramic Images

• Main Authors: Chen, Shih-Lun, Chen, Tsung-Yi, Mao, Yi-Cheng, Lin, Szu-Yin, Huang, Ya-Yun, Chen, Chiung-An, Lin, Yuan-Jin, Hsu, Yo-Ming, Li, Chi-An, Chiang, Wei-Yuan, Wong, Kai-Yi, Abu, Patricia Angela R
• Format: text
• Published: Archīum Ateneo 2022
• Subjects: Teeth; Dentistry; Image segmentation; Urban areas; Training; Sensors; Implants; Convolutional neural network (CNN); image enhancement; image segmentation; Panoramic (PANO); tooth disease recognition; Computer Sciences; Physical Sciences and Mathematics
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/336; https://doi.org/10.1109/JSEN.2022.3211981
• Institution: Ateneo De Manila University

For a daily dental practice, the Panoramic (PANO) X-ray film is one of the most commonly used dental X-rays. One of its important advantages is the coverage of most anatomic structures and clinical findings in a single film. Important information about clinical treatment and diagnosis can be provided from the expert analysis of the PANO. Combined with the assistance of artificial intelligence, the application has great potential. The purpose of this study was to propose an automated detection system based on several modern convolutional neural networks (CNNs) for the classification of retained roots, endodontic treated teeth, and implants. In order to meet the standards of practical clinical application, the database used in this study is provided by dentists with more than three years of practical experience. The contributions of this work are given as follows: 1) proposed more advanced techniques for image segmentation and image position in dental radiographs; 2) a better image enhancement is proposed, which improves the accuracy of the five CNNs to more than 96%; and 3) combined with the fuzzy operation to achieve more powerful and accurate anomaly detection. The final result has an accuracy rate of up to 98.75%. It is about 20% higher than previous techniques. This research designed to identify and document each specific finding automatically could help dentists obtain an objective treatment evaluation and provide dentists more precious clinical time for dental operations and communication with patients.