Single-model deep learning approach for simultaneous cervical vertebral maturation staging and skeletal jaw relationship on lateral cephalograms using YOLOv8 and CNN.

backgroundThis study aimed to develop an integrated artificial intelligence (AI) pipeline for cervical vertebral maturation (CVM) staging and skeletal jaw relationships and to validate its output against the results of human observers.

methodsA total of 720 lateral cephalograms were collected from the archives of the orthodontic department. The participants' ages ranged from 8 to 18 years. Cephalograms were categorized into six cervical stages based on McNamara's criteria and classified into skeletal Class I, II, and III patterns based on their ANB angles. The dataset was divided into a training set (n = 540) and a test set (n = 180). The training set was used to train a convolutional neural network (CNN) and a YOLOv8 model. The test set of cephalograms was coded and randomly assigned to two orthodontists for comparison with the AI model results using weighted kappa and Cohen's kappa statistical analyses to verify accuracy.

resultsFor human observers, the interobserver agreement ranges were as follows: (κw ≈ 0.976-0.989) for skeletal classification and (κw ≈ 0.956-0.999) for CVM staging, while the intra-observer reliability was also almost perfect for both methods (κw ≈ 0.82- 0.85). Substantial agreement was found between the generated AI model and human observers for skeletal classification (κ = 0.725) and CVM staging (κ = 0.786). Both results were statistically significant (p < 0.01).

conclusionsWithin the limitations of this study, the AI model exhibited substantial agreement with human observers for CVM staging and skeletal assessment, demonstrating its potential viability as a clinical decision-support tool.