B-Gn-F plane proves as the most accurate best mandibular midsagittal plane to segment and identify mandible anatomical landmarks

According to a recent study, deep learning can segment and identify mandibular midline landmarks with the help of the best mandibular midsagittal plane (MMSP), which is B-Gn-F. This study was published in the European Journal of Medical Research.

This study included 400 participants who were randomly assigned into a training group (n=360) and a validation group (n=40), along with a test group (n=50) comprising normal individuals. PoseNet identified 27 anatomic landmarks from the mandible images from cone-beam computed tomography (CBCT) using the PointRend deep learning mechanism. For the test group, three-dimensional (3D) coordinates were obtained for five central landmarks and two pairs of side landmarks. Using the template mapping technique, every 35 combinations of 3 midline landmarks were screened. For each of the 35 mirror planes, the asymmetry index (AI) was calculated. The top 4 planes having the smallest AIs were compared through volume difference, similarity index, and distance in order to find the plane with the least errors.

At the end of the study, it was found that in 10 ± 1.5 seconds, the mandible was automatically segmented with a 0.98 Dice similarity coefficient. Among the 27 landmarks, the average localization error was 1.04 ± 0.28 mm. The plane created by B (supramentale), Gn (gnathion), and F (mandibular foramen) should be used by MMSP. The average AI grade observed was 1.6. No significant difference was seen in volume or distance, although the similarity index was significantly different.

From the above results, it can be concluded that deep learning may segment and identify mandibular midline landmarks with the help of the best mandibular midsagittal plane (MMSP), which is B-Gn-F.