An artificial intelligence-based screening tool for orthognathic surgery using MKG angle in lateral cephalograms
Issued Date
2025-04-01
Resource Type
ISSN
23214600
eISSN
23211407
Scopus ID
2-s2.0-105007213451
Journal Title
Apos Trends in Orthodontics
Volume
15
Issue
2
Start Page
134
End Page
140
Rights Holder(s)
SCOPUS
Bibliographic Citation
Apos Trends in Orthodontics Vol.15 No.2 (2025) , 134-140
Suggested Citation
Limcharoen N., Thiradilok S., Thanathornwong B., Manopatanakul S. An artificial intelligence-based screening tool for orthognathic surgery using MKG angle in lateral cephalograms. Apos Trends in Orthodontics Vol.15 No.2 (2025) , 134-140. 140. doi:10.25259/APOS_153_2024 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/110606
Title
An artificial intelligence-based screening tool for orthognathic surgery using MKG angle in lateral cephalograms
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Objectives: This study aimed to develop a clinical decision support system utilizing the MKG angle – derived from points M, K, and G – as a novel neural network parameter for evaluating sagittal maxillo-mandibular discrepancy. This system serves as a pre-operative screening tool for predicting the need for orthognathic surgery. Material and Methods: This retrospective study collected 494 digital lateral cephalograms. MKG angle values extracted from these cephalograms were analyzed using a Keypoint Region-based Convolutional Neural Network integrated with Detectron2 for object detection. Analysis was conducted using Keras software to facilitate decision-making regarding orthognathic surgery. The model’s output ranged from 0 to 1, with values closer to 1 indicating a stronger recommendation for orthognathic surgery. A training loss graph was used to monitor the model’s performance over epochs, while a confusion matrix evaluated the model’s accuracy and predictive capabilities. Results: The training loss value for the object detection model was 3.0510. Model performance was further evaluated using metrics such as root mean square error (RMSE) and percentage of detected joints (PDJ). The RMSE was measured at 2.68 pixels, while the PDJ, with a threshold of 0.05, achieved a value of 0.99, indicating a high level of accuracy. The developed system achieved an orthognathic surgery diagnosis accuracy of 70.41%, with a training loss value of 0.6163. The evaluation revealed instances of misdiagnosis; out of 98 cases, 29 were identified as misdiagnosed through a confusion matrix. The model’s sensitivity and specificity were measured at 72.5% and 68.97%, respectively. Conclusion: A supplementary tool for orthognathic screening, utilizing two-dimensional digital lateral cephalometry images and MKG angle as a parameter, was developed by merging a neural network model with clinical decision-making.