IADR Abstract Archives
Prediction of Cervical Vertebral Maturation Stage: a Machine Learning Study
Objectives: Cervical vertebral maturation (CVM) is widely used to evaluate growth potential in the field of orthodontics. The aim of this study is to develop an artificial intelligence (AI) algorithm to automatically predict and classify the CVM stages in terms of growth phases (pre-, circum-, and postpubertal) using the cone-beam computed tomographic (CBCT) images.
Methods: A total of 30,016 slices obtained from 56 patients with the age range of 7-16 years were included. After cropping the region of interest (ROI), a convolutional neural network (CNN) was built to classify the slices based on the presence of a good vision of vertebrae for classification. The output was used to train another model capable of categorizing the slices into phases of growth, which were defined as Phase 1 (prepubertal, CVM stages 1 and 2), phase 2 (circumpubertal, CVM stage 3), and phase 3 (postpubertal, CVM stages 4, 5, and 6). After training the model, 88 unused images belonging to 3 phases were used as the testing dataset to evaluate the performance of the model using multi-class classification metrics.
Results: The average classification accuracy of the first and second CNN-based deep learning models were 96.06% and 94.74%, respectively on the validation dataset. The multi-class classification metrics applied to the testing dataset also showed an overall accuracy of 84% for predicting the growth phases. Moreover, phase I ranked the highest accuracy in terms of F1 score (87%), followed by phase II (83%), and phase III (80%).
Conclusions: Our proposed models could automatically detect the good visions of C2-C4 vertebrae required for CVM staging and accurately classify slices into 3 growth phases without the need for annotating the shape and configuration of vertebrae. This will result in developing a fully automatic and less complex system with reasonable performance, comparable to expert practitioners.
Methods: A total of 30,016 slices obtained from 56 patients with the age range of 7-16 years were included. After cropping the region of interest (ROI), a convolutional neural network (CNN) was built to classify the slices based on the presence of a good vision of vertebrae for classification. The output was used to train another model capable of categorizing the slices into phases of growth, which were defined as Phase 1 (prepubertal, CVM stages 1 and 2), phase 2 (circumpubertal, CVM stage 3), and phase 3 (postpubertal, CVM stages 4, 5, and 6). After training the model, 88 unused images belonging to 3 phases were used as the testing dataset to evaluate the performance of the model using multi-class classification metrics.
Results: The average classification accuracy of the first and second CNN-based deep learning models were 96.06% and 94.74%, respectively on the validation dataset. The multi-class classification metrics applied to the testing dataset also showed an overall accuracy of 84% for predicting the growth phases. Moreover, phase I ranked the highest accuracy in terms of F1 score (87%), followed by phase II (83%), and phase III (80%).
Conclusions: Our proposed models could automatically detect the good visions of C2-C4 vertebrae required for CVM staging and accurately classify slices into 3 growth phases without the need for annotating the shape and configuration of vertebrae. This will result in developing a fully automatic and less complex system with reasonable performance, comparable to expert practitioners.