Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Automatic dental age assessment using deep learning contains 3 major steps: 1. Extraction of region of interest (ROI) from dental radiographs 2. Classification of these extracted ROIs into developmental stages of maturity and 3. Estimation of age based on predicted developmental stage. This thesis focuses on the second step of automatic dental age assessment system and evaluates three major hypothesis. First, the feasibility of a combined classification model for the second step of automatic dental age assessment system is investigated. The performance of the combined model for second and third molar is compared against the individual classification models of each teeth. The performance was evaluated using 5-fold cross validation and DenseNet201 network architecture gave the best performance in each case. The best combined classification model resulted in accuracy of 0.694, mean absolute error of 0.353, and linearly weighted kappa of 0.883. The performance of the combined model was significantly better than individual classification model of third molar, but slightly worse than second molar classification model. Next, the efficiency of an automatic ROI extraction technique is evaluated for the second step of automatic dental age assessment system. It was found that the classification models built using automatically extracted ROIs perform as good as the classification models built using manually extracted ROIs. Finally, the impact of extracted ROI size on the performance of classification models in the second step of automatic dental age system was evaluated. It was found that the performance of classification models decreases with the selection of extremely large sized ROIs. It is also suggested that the selection of ROI size should be done experimentally after evaluating the performance of classification model for various ROI sizes.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

The biological age of a person is critical in many legal settings related to sports and forensics. A typical scenario is to determine the age of athletes, unaccompanied fugitives claiming to be minors or a corpse. This demands for a method to determine the biological age of a person by doing relevant tests. In this regard, third molar-based age estimation looks a promising approach as the third molar shows distinct patterns of growth corresponding to the biological age of the person. The current methodology of age estimation from third molars involves radiology expertise to manually categorize the panoramic X-ray images to the development stage and later use statistical models to predict the biological age. This process is time-consuming and cumbersome. The goal of the project is to automate the age assessment process from dental radiograph images. The focus of the thesis is on exploring the advancements in deep learning for dental age estimation with an emphasis on regression and ordinal regression methods. The Dataset for the experiments consists of 2214 dental radiograph images distributed over age ranges of 8-24 years. Five-fold cross-validation was used to evaluate the models. A deep learning model with, Densenet201 as the convolutional neural network, achieved the best results with the ordinal regression method. The corresponding Mean Absolute Error (MAE) was 1.15 years.