Measuring Adult Facial Morphology in Three Dimensions
Abstract
Objectives: The aim of this study is to evaluate the reliability of measuring three-dimensional soft tissue morphology using a laser imaging system. Design: Prospective clinical trial.
Materials and Methods: Thirty-eight adult subjects, mean age 24.5 years, were analyzed for soft tissue changes at baseline (T1) and at 1 week (T2) using two commercially available Minolta Vivid 900 (Osaka, Japan) laser scanning devices assembled as a stereopair. Left and right images were merged to form the whole face, and these images were superimposed to assess the errors between the two faces at T1 and at T2.
Results: The results showed that the mean shell deviations for left and right scans at T1 were 0.32 ± 0.08 mm and 0.30 ± 0.09 mm for males and females, respectively. The mean shell deviations for left and right scans at T2 were 0.34 ± 0.08 mm and 0.32 ± 0.09 mm for males and females, respectively. The mean difference of the merged composite faces superimposed at T1 and T2 was 0.37 ± 0.07 mm and 0.35 ± 0.09 mm for males and females, respectively. Paired t-tests revealed that the mean difference of 0.02 mm was statistically insignificant (P > .05). The reproducibility error was 0.7 and 0.8 mm for females and males, respectively, when a tolerance of 90% was imposed on the aligned faces.
Conclusions: Capturing soft tissue morphology of the face, using the technique described, is clinically reproducible within 1 week of the initial records.
INTRODUCTION
The world of three-dimensional (3D) technology has developed at a rapid pace over the last decade, allowing newer machines and advanced software support to be created. These advancements have produced fast, efficient, and cost-effective applications for the clinical and laboratory settings. Likewise, medical imaging has moved from two-dimensional representations (radiographs and color photographs) to more sophisticated 3D techniques. At present, one is able to recreate the facial form to a greater precision using these sophisticated tools.
The laser scanner can be used as a soft tissue scanner and is part of an array of imaging devices for the creation of 3D images. Images using this technique have been created to establish databases for normative populations,1 for cross-sectional growth changes,23 and for the assessment of clinical outcomes in surgical4–11 and nonsurgical treatments12–14 in the head and neck region.
Assessing the accuracy of soft tissue simulation, however, is complex.15 All systems are affected by changes in muscular tone, nasal breathing, and head posture of the subjects measured. Previous reports have found that facial morphology was reproducible16 and feasible on children.17 However, to date, no study has effectively quantified the errors associated with the capture of 3D adult morphology.
Previous studies have reported on the validity and high accuracy of the Minolta 700 and 900 scanners and found them to be accurate to the level of 1.9 ± 0.8 mm18 and 1.1 ± 0.3 mm.10 Independent studies by the authors show that the Minolta 900 is accurate to a level of 0.56 ± 0.25 mm and the error in computerized registration of left and right scans is 0.13 ± 0.18 mm.19
With the validity of the scanning system already evaluated, this study aims to quantify the reproducibility of obtaining 3D laser scans over time for adults.
MATERIALS AND METHODS
Subjects
A cohort of adult dental students with normal facial proportions was invited to participate in this longitudinal study conducted at the 3D imaging laboratory in the dental hospital at Cardiff University. Approval for this study was obtained as part of a larger facial growth project from the participating subjects and the relevant ethics committee. In addition, positive written consent was obtained from each subject who participated in the project.
Three-dimensional Imaging System
The laser scanning system consisted of two high-resolution Minolta Vivid VI900 3D cameras, with a reported manufacturing accuracy of 0.1 mm, operating as a stereopair. Each of these cameras emits an eye-safe Class I laser (FDA) λ = 690 nm at 30 mW with an object-to-scanner distance of 600 to 2500 mm and a fast mode scan time of 0.3 seconds. The system uses a one-half-frame transfer charged-coupled device and can acquire 307,000 data points. The scanner's output data is 640 × 480 pixels for 3D and red, green, and blue color data. Data were recorded on a desktop workstation with a two GHz Pentium 4 processor. For surface registration, a Minolta medium-range lens with a focal length of 14.5 mm was used. The cameras were placed at a distance of 1350 mm from the subjects. The scanners were controlled with Multi-scan™ software (Cebas Computer GmBH, Eppelheim, Germany), and data coordinates were saved in a vivid file format (vvd).
Information was transferred to a reverse modeling software package Rapidform™ 2004–RF4 (INUS Technology Inc, Seoul, Korea)—RF4 for analysis. This software provides nine different 3D work activities and together allows high-quality polygon meshes, accurate freeform Non-Uniform Rationale B-Spline surfaces, and geometrically perfect solid models to be created. RF4 generates data as absolute mean shell deviations, standard deviations of the errors during shell overlaps, maximum and minimum range maps, histogram plots, and finally color maps. All linear measurements were made in millimeters.
Data Capture Technique
A custom-made portable studio facilitated standardized light conditions. The studio was sufficiently compact to fit into a corner of a classroom or medical room without difficulty and house all the necessary equipment. Natural Head Posture (NHP) was adopted for this study because this has been shown to be clinically reproducible.20
The subjects sat on a self-adjustable stool and were asked to look into a mirror with standard horizontal and vertical lines simulating a cross marked on it. They were asked to level their eyes to the horizontal line, and the midline of the face was aligned to the vertical line. Adjustments to seating heights were made to assist the subjects in achieving NHP. The subjects were also instructed to swallow hard and to keep their jaws in a relaxed position just before the scans were taken. The scans were taken at the same time interval, and the total scan time was approximately 7.5 seconds. If it was perceived that the subjects moved between scans, the procedure was repeated. One raw data set, comprising one left and right laser scan, was taken of each subject. The scans were taken at baseline (T1) and at one week (T2).
Data Processing of Left and Right Facial Scans
Extraneous data were removed by an in-house developed software subroutine, which took 30 seconds to complete. This automatic and systematic process further reduced the scanned images into shells and identified those small shells that represented minor scanning distortions. These images were smoothed out, preserving all shape and volume, and the left and right scans were aligned to one another on the basis of the areas of overlap of the faces. The premerged scans were carefully checked individually and unwanted areas that could not be automatically removed were done so manually by dividing the unwanted areas from the main shell before proceeding to the next stage. Finally, one composite whole face, per individual subject, for every time frame was generated.
Data Processing of the Whole Faces
Individual whole faces of subjects were superimposed over one another to determine changes that occurred at T1 and at T2. This systematic process starts by manually aligning the five points on the facial scans (four points at the outer and inner cantus of the eyes and one point on the nasal tip) and subsequently by fine registration where the computers determined the best-fit of the two scans. Further details may be obtained from previous referenced sources.1721
Level of Tolerance
To obtain an overall clinical picture, colored face maps were generated to determine the patterns within the face where the error was considered to be high. Tolerance levels were set for shell deviations and calculated automatically by the software. Any deviations between the faces during the two time intervals that corresponded to a tolerance level above a certain threshold were shown in color and any values below the tolerance interval showed up in black. Incremental levels of 0.1 mm were analyzed from 0.5 to 0.9 mm. This helped determine the reproducibility of the face over the time frames T1 and T2.
Statistical Analyses
Within RF4, a shell-to-shell deviation map was computed and automatically produced. The results include the maximum and minimum range of shell deviations, the average distance between the two shells, and the standard deviation. This function was used to statistically analyze the mean shell deviations and standard deviations for left and right premerged scans and also the differences in whole face soft tissue morphology between the merged faces over the two time frames.
The mean shell deviations were tested for normality and differences between the groups measured were analyzed using the Student's t-test (SPSS, Chicago, Ill). P values less than .05 were considered significant. This was undertaken for premerged left and right scans and also for the whole faces superimposed over one another at T1 and at T2.
RESULTS
Thirty-eight selected individuals (19 males and 19 females with a mean age of 24.5 years) were recruited to participate in this study.
Reliability of Merging Left and Right Scans
The data set for each subject is represented diagrammatically in Figures 1 and 2. The results showed that the mean shell deviations for scans at T1 were 0.32 ± 0.08 mm and 0.30 ± 0.09 mm for males and females, respectively. The mean shell deviations for the scans at T2 were 0.34 ± 0.08 mm and 0.32 ± 0.09 mm for males and females, respectively. Each of these data sets was tested for normality and was found to be normally distributed. Paired t-tests carried out on the mean shell deviations between T1 and T2 revealed no significant differences between sexes (P > .05). The mean shell deviation of the left and right scans before merging for time intervals T1 and T2 are summarized in Table 1.
Citation: The Angle Orthodontist 76, 5; 10.1043/0003-3219(2006)076[0773:MAFMIT]2.0.CO;2
Citation: The Angle Orthodontist 76, 5; 10.1043/0003-3219(2006)076[0773:MAFMIT]2.0.CO;2
Reliability of Whole Face as T1 and T2
The mean shell deviations between superimposed whole faces are shown in Table 2. The results showed that the mean difference of the merged composite face for T1 and T2 was 0.37 ± 0.07 mm and 0.35 ± 0.09 mm for males and females, respectively. The difference between the two sexes was 0.02 ± 0.05 mm. Paired t-tests were carried out on mean shell deviations (P = .08) and were not significant.
Level of Tolerance
The amount of overlap between two faces, expressed in percentages for the tolerance levels at increments of 0.1 mm from 0.5 to 0.9 mm, are shown in Figure 3. In general, if the clinical difference was seen in less than 90% of the face, this was deemed to be reliable and reproducible. Aligned facial maps of the merged scans (T1 and T2) showed that on average 90% of the created composite facial scans correlated to one another with an error up to 0.8 for males and 0.7 mm for females. These were considered to be a clinically acceptable level of tolerance. The results are summarized in Table 3.
Citation: The Angle Orthodontist 76, 5; 10.1043/0003-3219(2006)076[0773:MAFMIT]2.0.CO;2
DISCUSSION
Most studies have concentrated on reliably measuring distances between chosen anthropometrics points on the 3D-generated images against corresponding points on live subjects522–24 as a form of validation. Some studies use complex mathematics to derive and analyze shapes.2526 Recently, attempts have been made to analyze the dynamic face by linear measurement between points27 and facial polygons.28
Error studies to measure accurate facial soft tissue reproducibility are rare. Only two studies to date have attempted this. The first study used a small sample consisting of adults and the images were averaged before measuring between time frames.2 This potentially amalgamates all errors during the averaging process (for example, cancellation of positive and negative errors) and may not give a true picture of reproducibility. The second study was carried out on children and found that the average error was 0.85 mm over 3 days.16 The result of our study closely relate to those of the second study and is important in eliminating systematic error associated with imaging subjects, and we hope it creates a more accurate clinical analysis of treatments and growth changes. For example, when a reproducibility error of 0.7 mm is applied to superimposed images of a patient undergoing surgery, a more realistic assessment of the treatment changes may be made (Figure 4). The facial changes correspond well to the type of surgery performed when a tolerance is placed. These facial areas and shape changes show a forward movement of the maxilla and a backward movement of the mandible.
Citation: The Angle Orthodontist 76, 5; 10.1043/0003-3219(2006)076[0773:MAFMIT]2.0.CO;2
All imaging systems have some form of error associated with their application. It is important that clinicians understand the potential limitations and investigate the errors associated with the techniques. The future for 3D imaging is promising. The use of this laser scanning technique has shown potential in assessing changes in facial morphology as a result of orthodontic treatment, surgery, and facial growth.
CONCLUSIONS
-
Using RF4, the mean shell deviation in superimposition of whole faces was 0.37 ± 0.07 mm for males and 0.35 ± 0.09 mm for females, as shown by scans taken within 1 week.
-
The mean difference in facial morphology between males and females was 0.02 ± 0.08 mm and was not clinically significant.
-
The reproduction of 90% of facial morphology is accurate to within 0.7 mm for females and 0.8 mm for males.
-
The 3D images may be used as accurate representations of facial morphology within the errors reported.
Shell deviation scores for males at T1 and at T2
Shell deviation scores for females at T1 and at T2
The facial maps of a female individual when tolerance values of 0.5, 0.7, and 0.9 mm were applied to the T1 and T2 faces. There was an improvement in facial overlap and the values were 84.5%, 93.6%, and 97.1%, respectively. These percentages indicate that there was no change in facial pose at the relevant tolerance values.
Facial images of a patient who underwent a bimaxillary orthognathic surgical procedure to correct a Class III facial abnormality. The associated changes may be seen (red, positive; blue, negative) when a tolerance reproducibility error of 0.7 mm is applied to the superimposed shells. The resultant soft tissue changes as a result of surgery can be more accurately mapped in three-dimensions
Contributor Notes
Corresponding author: Dr. Chung How Kau, Wales College of Medicine, Biology, Health and Life Sciences, Dental Health and Biological Sciences, Heath Park, Cardiff, Wales CF14 4XY, UK (kauc@cardiff.ac.uk)