Spatial Configuration-Net Aided Spheno-Occipital Synchondrosis Ossification and Staging in Orthodontics

Abstract

Aim or Purpose: To stage spheno-occipital synchondrosis (SOS) ossification using Cone Beam CT(CBCT) first, followed by the application of AI-based modified Convolutional Neural Networks for the automation of tasks. Materials and Methods: We used IRB exempted protocol (#830154) to analyze CBCTs obtained for 69 male and 67 female patients between the ages 5 and 25 using Osirix® and ITK Snap® software. Mid-sagittal sections of maxillofacial CBCT were obtained in order to visualize the SOS. Images were labeled according to chronological age in years and months and sorted based on gender. Spatial Configuration Net (SCN) was used to make landmark predictions. Using multiple hand-staged images of the SOS, the Euclidean pixel-wise distance between landmarks for a given stage was calculated and normalized. A support Vector Machine (SVM) algorithm was used to predict the stage of SOS ossification. Results: The SVM-classified SOS stages were compared with human classifications to see the level of accuracy using the Sorensen-DICE index. SOS ossification started around 9.3 years in males and 8.7 years in females, and completely fused by the age of 17.8 for males and 14.9 for females. Conclusions: SOS is a stable skeletal landmark and using our novel staging system proposed here, accurate determination of the skeletal age of a patient is possible. Manual identification of the ossification at SOS may lead to errors in age estimation, but artificial intelligence-based automation might be a useful clinical tool for orthodontists in the estimation of skeletal age.