Automated Measurement of Midpalatal Suture Density Ratio Based on Deep Learning

Abstract

Rapid maxillary expansion (RME) is a common method for maxillary transverse deficiency. While the midpalatal suture density (MPSD) ratio is a critical predictor of skeletal responsiveness to RME, its manual measurement on cone-beam computed tomography (CBCT) scans is time-consuming and prone to inaccuracies in delineating the regions of interest. This study aimed to establish a deep learning-based automated MPSD quantification method to enable rapid and standardized evaluation. A total of 400 CBCT scans were collected. Image slices containing the midpalatal suture, hard palate, and soft palate regions were extracted to create segmentation datasets. Three deep learning networks (U-Net, Deeplab v3, and Segformer) were trained, with the optimal model selected for automated segmentation. Gray values from the model-predicted regions were automatedly quantified, and the MPSD ratio was calculated using a standardized formula. Agreement between automated and manual MPSD ratio calculations was assessed by using a paired t-test. U-Net achieved superior segmentation accuracy with F1-scores of 0.845 (midpalatal suture), 0.854 (hard palate), and 0.911 (soft palate), respectively. The proposed system completed MPSD ratio measurements in 1 min per case, 3 times faster than manual methods. The paired t-test showed no statistically significant difference between automated and manual measurements. The deep learning solution transforms MPSD ratio measurement from a time-consuming process into an efficient, automated workflow, accelerating treatment decision-making.

Keywords: Deep learning; Image segmentation; Midpalatal suture density ratio; Rapid maxillary expansion.