Witsarut Upalananda. Semi-automated technique to assess the Developmental stage of mandibular third molars for age estimation. Master's Degree(Dentistry). Chiang Mai University. Library. : Chiang Mai University, 2568.
Semi-automated technique to assess the Developmental stage of mandibular third molars for age estimation
Abstract:
Background: Dental age estimation is a method for estimating the age of people with unknown date of birth, a requirement for age-related judgment in legal procedures. However, the procedure for estimating dental age presents some difficulties. Training of observers to classify the stages of tooth development correctly or the use of experienced observers is needed to produce accurate and reproducible results. Application of artificial intelligence in dental age estimation makes the procedure of dental age estimation more reliable and less subjective. Objectives: This study aimed to develop a semi-automated technique to assess a developmental stage of mandibular third molars and test the accuracy of the technique. Materials and methods: This study focused on the last five developmental stages of the Demirjian et al. classification. Panoramic radiographs of 2,235 patients were collected. The radiographs were cropped manually to obtain isolated radiographic images of mandibular third molars. The samples in this study were 4,000 radiographic images of mandibular third molars, which consisted of 800 images of each developmental stage. The samples were assigned to training and test datasets. The training dataset had 3,600 images, which consisted of 720 images of each developmental stage, whereas the test dataset had 400 images, which consisted of 80 images of each developmental stage. A developmental stage assessment technique was developed using two deep learning algorithms, the AlexNet and GoogLeNet models. The algorithms were trained using the training dataset and their accuracy was tested using the test dataset. This study evaluated intra- and inter-observer reliability using Cohens kappa coefficient. Accuracy of the developed techniques was evaluated using the percentage of correction. Results: Cohens kappa coefficients of intra- and inter-observer reliability were 0.898 and 0.833, respectively. The AlexNet model was 79.75% accurate in assessing the developmental stage of mandibular third molars, whereas the GoogLeNet model had 82.50% accuracy. The misclassified results of the AlexNet model deviated one to two stages from human-assessed results, whereas the misclassified results of the GoogleNet model deviated only one stage. Conclusions: This semi-automated technique was highly accurate in assessing the development of the mandibular third molar. The GoogLeNet model is recommended for the task because it has higher accuracy than the AlexNet model. Dentists can apply this developed technique to assist in the process of age estimation of adolescents and young adults.