Automatic 3D dense phenotyping provides reliable and accurate shape quantification of the human mandible

Abstract

Automatic craniomaxillofacial (CMF) three dimensional (3D) dense phenotyping promises quantification of the complete CMF shape compared to the limiting use of sparse landmarks in classical phenotyping. This study assesses the accuracy and reliability of this new approach on the human mandible. Classic and automatic phenotyping techniques were applied on 30 unaltered and 20 operated human mandibles. Seven observers indicated 26 anatomical landmarks on each mandible three times. All mandibles were subjected to three rounds of automatic phenotyping using Meshmonk. The toolbox performed non-rigid surface registration of a template mandibular mesh consisting of 17,415 quasi landmarks on each target mandible and the quasi landmarks corresponding to the 26 anatomical locations of interest were identified. Repeated-measures reliability was assessed using root mean square (RMS) distances of repeated landmark indications to their centroid. Automatic phenotyping showed very low RMS distances confirming excellent repeated-measures reliability. The average Euclidean distance between manual and corresponding automatic landmarks was 1.40 mm for the unaltered and 1.76 mm for the operated sample. Centroid sizes from the automatic and manual shape configurations were highly similar with intraclass correlation coefficients (ICC) of > 0.99. Reproducibility coefficients for centroid size were < 2 mm, accounting for < 1% of the total variability of the centroid size of the mandibles in this sample. ICC’s for the multivariate set of 325 interlandmark distances were all > 0.90 indicating again high similarity between shapes quantified by classic or automatic phenotyping. Combined, these findings established high accuracy and repeated-measures reliability of the automatic approach. 3D dense CMF phenotyping of the human mandible using the Meshmonk toolbox introduces a novel improvement in quantifying CMF shape.

Automated 3D Landmarking of the Skull: A Novel Approach for Craniofacial Analysis

“…Only one of the observers was trained in cranial landmarking, resulting in a significantly lower overall RMS error (0.66mm) in comparison to the other two observers (1.11mm and 0.94mm). These values are similar to those reported in other studies [27, 30, 31], albeit slightly higher than findings from a study employing specialized landmarking software at some landmarks [31]. As expected, the inter-observer error was higher (1.44mm).…”

“…Analysis revealed a variation in Euclidean distance between manual and automatic landmarks ranging from 0.10mm - 7.24mm, with an average of 1.5mm. Notably, landmarks exhibiting higher values were often associated with regions that were less effectively masked (e.g., gonion where the mask did not always align with the edge of the skull) and were prone to higher inter-observer errors, as observed in previous studies [30]. Previous research has reported comparable errors between manual and automatic landmark placements ranging from 2.19mm [31], 1.4mm [30], 2.01mm [42] and 1.26mm [27].…”

“…Consequently, it is plausible that this issue may be more attributable to limitations inherent to the CBCT imaging technique rather than deficiencies in the masking methodology. Notably, the maximum Meshmonk error is higher than that seen for a mandible mask [30], albeit confined to these specific regions. Our errors were reduced by modifying the Meshmonk settings “NRM.FlagFloatingBoundary” and “NRM.FlagTargetBoundary” (for CT both were set to true, for CBCT only the first was set to true) and recommend that these settings be tested in combination with our script to produce a type of ‘quality control’ test image that depicts the mask overlaying the original mesh (see Supplementary File) to define the best settings for each user’s imagery.…”

“…While its application has led to multiple publications exploring the genetic architecture of the human face [2,3,6], it is crucial to note however, that the framework does not include a complete hard-tissue component. Global registration masks using Meshmonk have been developed and utilized for specific segments of the skull, including the lower jaw [30] and the cranial vault [9], however, assessment of the facial bones within the craniofacial complex has not been explored. The frontal aspect of the skulls facial skeleton poses significant challenges due to its intricate structure, with a separate lower jaw, as well as orbital and nasal cavities.…”

Shape variation and sexual dimorphism of the adult human mandible evaluated by geometric morphometrics

“…The first approach performs a rigid alignment of the template mesh to match the target, followed by deformable registration to refine the match, then transfers the landmarks from the atlas to the target mesh. Examples are ALPACA [5, 6] and MeshMonk [7, 8], which mainly differ in their non-rigid registration algorithm. Variants of this approach abolish landmarks altogether and achieve correspondences between the vertices of the meshes directly [3, 9].…”