IADR Abstract Archives
An Artificial Intelligence (AI) model for the Segmentation of Teeth on Orthopantomograms (OPGs)
Objectives: To compare the accuracy of teeth identification on OPG using a Deep Learning (DL) AI algorithm model with human annotators for instance segmentation and teeth numbering.
Methods: OPGs were manually annotated by human annotators to lay down the ground truth for training two Convolutional Neural Network (CNN) algorithms, namely U-net and Faster RCNN. These algorithms were concurrently trained and validated on a dataset of 40 labelled OPGs. The U-net algorithm was trained on OPGs specifically annotated with fluid margins to label all 32 teeth via instance segmentation allowing each tooth to be denoted as a separate entity (figure 1). Simultaneously, the teeth were also numbered as per the FDI (Fédération Dentaire Internationale) system, using bounding boxes to train the Faster RCNN algorithm (figure 2). Consequently, both trained algorithms were combined to develop an AI model capable of segmenting and numbering all teeth on an OPG.
Results: The performance metrics of the algorithms were assessed relative to the ground truth laid down by human annotators in the training and validating datasets of OPGs. The performance of the U-net algorithm was determined using accuracy and dice coefficient values of 88% and 92%, respectively. The performance of the Faster RCNN algorithm was determined by the overlap accuracy being 31.19 bounding boxes (out of a possible of 32 boxes); the classifier accuracy of labels was 98%.
Conclusions: The ability of an AI model to automatically identify teeth on OPGs will aid dentists with diagnosis and treatment planning. This will lead to the reduction of workload and diagnosis time of dental professionals, eventually increasing the efficiency as well as accuracy of dental treatment. The instance segmentation and teeth numbering results of our trained AI model were exceptionally close to the ground truth; holding a promising future for its incorporation into clinical dental practice.
Methods: OPGs were manually annotated by human annotators to lay down the ground truth for training two Convolutional Neural Network (CNN) algorithms, namely U-net and Faster RCNN. These algorithms were concurrently trained and validated on a dataset of 40 labelled OPGs. The U-net algorithm was trained on OPGs specifically annotated with fluid margins to label all 32 teeth via instance segmentation allowing each tooth to be denoted as a separate entity (figure 1). Simultaneously, the teeth were also numbered as per the FDI (Fédération Dentaire Internationale) system, using bounding boxes to train the Faster RCNN algorithm (figure 2). Consequently, both trained algorithms were combined to develop an AI model capable of segmenting and numbering all teeth on an OPG.
Results: The performance metrics of the algorithms were assessed relative to the ground truth laid down by human annotators in the training and validating datasets of OPGs. The performance of the U-net algorithm was determined using accuracy and dice coefficient values of 88% and 92%, respectively. The performance of the Faster RCNN algorithm was determined by the overlap accuracy being 31.19 bounding boxes (out of a possible of 32 boxes); the classifier accuracy of labels was 98%.
Conclusions: The ability of an AI model to automatically identify teeth on OPGs will aid dentists with diagnosis and treatment planning. This will lead to the reduction of workload and diagnosis time of dental professionals, eventually increasing the efficiency as well as accuracy of dental treatment. The instance segmentation and teeth numbering results of our trained AI model were exceptionally close to the ground truth; holding a promising future for its incorporation into clinical dental practice.
3645137_File000010.jpg
3645137_File000011.jpg