Validity and reliability of masseter muscles segmentation from the transverse sections of Cone-Beam CT scans compared with MRI scans.

Abstract

To evaluate the validity and reliability of cone-beam computed tomography (CBCT) masseter muscle segmentation by comparing with the magnetic resonance imaging (MRI) masseter muscle segmentation of the same patients. Seventeen volunteers were included in this study. CBCT and MRI scans of the volunteers were taken, respectively, within one month. The masseter muscles in the CBCT scans were segmented by a generative adversarial network (GAN)-based framework combined with manual check. The masseter muscles in the MRI scans were segmented manually. The segmentations were repeated by the first examiner and a second examiner. For cross-sectional area (CSA), paired t-test, intraclass correlation coefficient (ICC) and standard error of measurement (SEM) were calculated to evaluate the validity and reliability of the segmentations. The validity and reliability were also calculated by Dice similarity coefficient (DSC) and average Hausdorff distance (aHD) between different segmentations. Seventeen volunteers were included in this study. CBCT and MRI scans of the volunteers were taken, respectively, within one month. The masseter muscles in the CBCT scans were segmented by a generative adversarial network (GAN)-based framework combined with manual check. The masseter muscles in the MRI scans were segmented manually. The segmentations were repeated by the first examiner and a second examiner. For cross-sectional area (CSA), paired t-test, intraclass correlation coefficient (ICC) and standard error of measurement (SEM) were calculated to evaluate the validity and reliability of the segmentations. The validity and reliability were also calculated by Dice similarity coefficient (DSC) and average Hausdorff distance (aHD) between different segmentations. Paired t-test showed that there was no significant difference in CSA between CBCT and MRI masseter segmentations. The ICCs were all larger than 0.95 and the SEM was less than 4.85 mm2 for CSA. The DSC was all larger than 0.95 showing over 95% of similarity between CBCT and MRI masseter segmentations. The aHD was all smaller than 0.09mm showing great consistency of the contour of CBCT and MRI segmentations. Masseter muscle segmentation from CBCT scans was not significantly different from the segmentation from MRI scans. CBCT muscle segmentation showed great validity compared with MRI scans, and great reliability in retests.