Application of a deep learning model to determine midpalatal suture maturation stage on CBCT

• Author / Creator

  Nik Ravesh, Mahshid

• Transverse maxillary deficiency is a condition characterized by a reduced transverse dimension of the upper jaw, commonly associated with posterior cross-bite, dental crowding, pharyngeal airway narrowing, and mouth breathing. Accurate staging of the mid-palatal suture (MPS) fusion is crucial for determining the appropriate treatment approach, whether surgical or non-surgical maxillary expansion. Traditionally, MPS staging is performed using cone beam computed tomography (CBCT), a technique that heavily relies on the practitioner's experience and is inherently subjective, leading to potential variability in assessment and treatment decisions.

  This study addresses these challenges by automating the classification and staging of MPS fusion through advanced deep learning (DL) techniques. We developed and trained both 2D and 3D convolutional neural network (CNN) models to enhance the accuracy, efficiency, and consistency of MPS evaluation. The 2D CNN model demonstrated remarkable performance with a high-test accuracy of 96.49% and excellent precision, recall, and F1-score values across all classification stages (AB, C, DE). This model highlights the effectiveness of traditional 2D approaches in handling MPS classification tasks.

  In contrast, the 3D CNN model, designed to capture the volumetric information of the MPS, achieved a test accuracy of 78.26%. Although this accuracy is lower compared to the 2D model, the 3D approach offers a more comprehensive evaluation by considering the full spatial context of the MPS, which could lead to more accurate staging in complex cases. The performance metrics for precision, recall, and F1-score in the 3D model were found to be acceptable, underscoring its potential for future refinement and optimization.

  The findings from this study underscore the potential of DL methods to revolutionize MPS fusion assessment by providing a more reliable, objective, and repeatable classification system. Such advancements could significantly enhance orthodontic treatment planning by offering clinicians a powerful diagnostic tool, improving patient outcomes, and reducing variability in treatment approaches. Furthermore, the techniques developed in this research have broader implications for medical image classification tasks beyond orthodontics, paving the way for the integration of AI-driven solutions in various medical imaging applications.

• Subjects / Keywords

  • Midpalatal
  • Suture
  • MPS
  • Deep Learning
  • CNN
  • Maturation

• Graduation date

  Fall 2024

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-2pjq-yx07

• License

  This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.