Although automated cephalometric landmark identification through artificial intelligence (AI) has long been attempted, its clinical application is now quickly approaching feasibility due to the development of deep learning models. In furtherance of this goal, our study created a regression-based convolutional neural network model, named AIPOD, that was trained on a larger and more variable set of data than typically seen in the literature. These modifications were made in hopes of improving the algorithm’s sensitivity and performance. Our study also created a new gold standard for measuring AI performance, relative to human performance, to evaluate the role of these programs in a clinical setting. The results will be presented in this lecture, as well as a demonstration of how the algorithm can be accessed and used from a web platform.
Learning Objectives:
After this session, attendees will be able to:
Explain how AIPOD contributes to the field through it's multi-center approach, large volume of training data, and establishment of a new gold standard to measure algorithm accuracy.
Show how AIPOD’s performance compares to landmark identification tasks completed by orthodontists.
Demonstrate how the algorithm can easily be deployed using a web interface.
