Sources of variation in the 3dMDface and Vectra H1 3D facial imaging systems

Abstract

Relative contributions to landmark variation. The experimental design employed is a complete randomized block design with repeated measures, thus the magnitude of variance within groups is attributable to differences made by the observer on the same object, or in our case the differences made by the MeshMonk toolbox in registering the same image multiple times (MeshMonk precision). This design also allows us to partition the magnitude of variance attributable to replicate images of the same person (participant error), and that attributable to the camera system (camera error). For these calculations we used a repeated measures ANOVA on the non-scaled and non-reflected GPA-aligned coordinates, with camera as a fixed effect and individual and replicate as random effects 40-42. For the set of 19 traditional landmarks, an ANOVA using the non-scaled and non-reflected GPA-aligned coordinates was similarly performed, with camera as a fixed effect and individual as a random effect.

Accuracy of Portable Face-Scanning Devices for Obtaining Three-Dimensional Face Models: A Systematic Review and Meta-Analysis

“…Among 21 articles, 15 papers had a low risk of bias [ 1 , 6 , 11 , 13 , 18 , 25 , 29 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 ], four articles displayed some level of unclear risk of bias [ 12 , 17 , 28 , 42 ], and two articles had a high risk of bias [ 15 , 43 ]. Considering applicability concerns, 13 articles exhibited a low level of concern [ 1 , 6 , 11 , 12 , 13 , 25 , 34 , 35 , 36 , 38 , 39 , 40 , 41 ], two articles showed an unclear level of concern [ 17 , 28 ], and six articles showed a high level of concern [ 15 , 18 , 29 , 37 , 42 , 43 ]. Major bias was found in the patient selection domain because some studies were not clear in method to provide random samples or used a small number of included volunteers/patients [ 11 , 15 , 18 , 34 , 41 , 42 , 43 ].…”

“…Impressions of patient faces were assessed in two studies [ 11 , 18 ] and human cadaver heads were assessed in one study [ 34 ]. Five studies [ 12 , 13 , 17 , 25 , 36 ] included both volunteers and mannequin heads but only the results drawn from the volunteers were considered within the scope of this review. In terms of the image acquisition technology, the 3D face-scanning systems used in all studies were classified into three major categories, namely, stereophotogrammetry, laser scanning, and structured light ( Supplementary Materials Table S2 ).…”

“…The mean discrepancy values of the face models obtained by portable facial scanners were below 1.0 mm, which is considered acceptable for clinical use [ 1 , 40 , 44 ]. Portable face scanners relying on laser or structured light technology showed more diversity than those that rely on stereophotogrammetry technology [ 11 , 12 , 13 , 18 , 25 , 28 , 29 , 42 , 43 ].…”

“…Overall, motion artifacts were considered the main source of error in the results of portable face-scanning systems [ 13 , 15 , 25 ], cautioning that the influence of involuntary facial movements has a greater impact on portable face-scan devices than stationary ones. Prolonged scanning time and unstable movements of the scanners may magnify the motion artifacts caused by involuntary facial movements [ 15 ].…”

“…To overcome the shortcomings of facial anthropometry and 2D face assessment techniques, three-dimensional (3D) optical scanning methods, such as stereophotogrammetry [ 11 , 12 , 13 , 14 ], laser scanning [ 1 , 15 , 16 ], and structure light scanning, were introduced [ 17 , 18 ]. These optical scanning methods provide 3D replication of the facial structure surface by generating a digital face model that can be coupled with a radiographic image of underlying bones and analytic software for 3D face analysis and virtual treatment planning or realistic surgery simulation [ 10 , 19 , 20 , 21 ].…”