Automated Classification of Midpalatal Suture Maturation Using 2D Convolutional Neural Networks on CBCT Scans

Mahshid Nik RaveshNazila AmeliManuel LagravereHollis Lai^*

• University of Alberta, Edmonton, Canada

Accurate assessment of midpalatal suture (MPS) maturation is critical in orthodontics, particularly for determining treatment strategies for maxillary transverse deficiency (MTD). Cone-beam computed tomography (CBCT) offers detailed imaging for MPS classification, but manual evaluation remains subjective and time-consuming. This study aimed to develop and evaluate a lightweight two-dimensional convolutional neural network (2D CNN) for automated classification of MPS maturation stages using axial CBCT images. A dataset of 580 images from 111 patients was retrospectively labelled according to Angileri's classification system and organized into three clinically actionable categories: AB (Stages A and B), C, and DE (Stages D and E). A CNN with nine convolutional layers was trained and evaluated using standard classification metrics and receiver operating characteristic (ROC) analysis. The model achieved a test accuracy of 96.49%, with class-wise F1-scores of 0.95 for AB, 1.00 for C, and 0.95 for DE. Corresponding AUC scores were 0.83 for AB, 0.84 for C, and 0.98 for DE. Despite strong classification performance, lower AUC values for early and transitional stages reflected known anatomical overlap and labeling subjectivity. These findings suggest that the proposed CNN is effective for MPS classification and may support real-time clinical decision-making in orthodontics, particularly in identifying cases requiring surgical intervention. Its compact architecture and high accuracy make it suitable for use in resource-limited environments, and future studies will explore volumetric models to further enhance diagnostic confidence.