Reproducibility of Novel Soft-Tissue Landmarks on Three-Dimensional Human Facial Scan Images in Caucasian and Asian

The study involved 80 European Caucasians (40 males, 40 females) and 80 Asian (40 males, 40 females). The age range of Caucasian volunteers was 20 to 50 years (30.49 ± 5.52 years). The age range of Asian volunteers was 18 to 45 years (30.36 ± 2.99 years). Written informed consent was obtained prior to enrollment. The study was approved by local university ethics committee (REF: 266-13) and conducted in accordance with the Declaration of Helsinki. Exclusion criteria were facial deformities, previous facial surgery, and volunteers diagnosed with epilepsy or other seizure disorders.

All facial 3d scans were captured by the high-resolution Vectra XT 3D Surface Imaging System (Canfield Inc., New Jersey, USA). It is a vertical fixed photography system with six integrated cameras at different angles. Its proprietary illumination system automatically adjusts the focus for optimal face imaging and its 3.5 mm photographic time reduces artifacts of the Vectra system due to the unconscious displacement of the subject.

Before taking the photograph, all volunteers were removed from any factors which would interfere with the image modeling: jewelry, glasses, and clothing elements (scarves, hat, etc.). They were asked to remove any hair from the face, forehead, and ears to completely expose the facial area. Male volunteers were asked to shave, while beards would cause image artifacts. The scan was performed in a well-lighted room. All volunteers were asked to sit in the same chair with a fixed backrest, lips kept closed without teeth grinding, and to look directly at the red marker dot on the 3D camera with a neutral facial expression and a natural head position. The lighting was kept under the same control conditions as in our daily work.

All captured images were processed, aligned, and analyzed using the proprietary Mirror® (Canfield Scientific; NJ, USA) [16]. The software was also implemented for providing reference frameworks with the x, y, z axis (sagittal, Y-Z plane; coronal, X-Y plane; and transverse, X-Z plane.) by unifying orientation for different images. The entire face was marked and symmetry of the planes was automatically adjusted by the integrated software. The midpoint between two endocanthions was chosen (mid-intercanthal point) as the origin point. The sagittal plane was referenced to the origin through the midline of the face, the coronal plane was determined to be the average natural head position, and the transverse plane was established to span the origin (Fig 1). The 3D face images were normalized on three planes to obtain comparable X, Y, and Z coordinates to assess the reproducibility of the facial landmarks [17].

In routine clinical facial surgery, surgeons will mark various anatomical landmarks for measurement. In this work, two raters identified and defined 46 landmarks based on former reports and the raters placed them separately on 3D facial images and all landmark coordinates were recorded (138 totally) for X,Y, Z axes [18‐20]. Their name, abbreviation, and definition are displayed in Table 1. These landmarks contain some of the most commonly used classic landmarks and the novel landmarks that we found beneficial for facial measurements (Fig 2). Both raters are professional researchers in our department experienced in 3D anthropometry. One of the raters evaluates the reproducibility of all landmarks in a two-week interval to obtain intra-rater reproducibility. We then compared the two raters' results for inter-rater reproducibility. Based on the intra-rater and inter-rater assessment results of 160 volunteer, we calculated the mean and standard deviation for each landmark. The error of the landmarks is presented as the absolute difference in measurement of each landmark on three axes. It is divided into three categories (<0.5 mm, <1 mm, and >1 mm). Coordinates with a difference between the measurements of less than 0.5 mm of all samples are classified as highly reproducible; with a difference between 0.5 mm and 1 mm are determined to be moderately reproducible; with a difference above 1 mm are considered to be poorly reproducible [21, 22]. Both intra- and inter-rater evaluations assessed the reproducibility of landmarks of all samples on three planes, divided into race and gender.

A total of 88320 variables (46 landmarks×160 subjects×3 planes×2 raters×2 measurements) were analyzed with SPSS Statistics 23.00 (IBM, Armonk, NY, USA). Based on the absolute difference of each landmark on the x, y, and z axes, we calculate the total average reproducibility difference using the following formula to: \(T=\sqrt{\frac{{(\Delta X)}^{2}+{(\Delta Y)}^{2}+{(\Delta Z)}^{2}}{3}}\), where T is the total average difference, \(\Delta X\) is the difference on the x-axis, \(\Delta Y\) is the difference on the y-axis, and \(\Delta Z\) is the difference on the z axis. Each variable is first corrected to the median for all volunteers. Bland–Altman plots were carried out for intra- and inter-rater reproducibility assessments. For each plot, the difference between the measurements of each landmark coordinate was calculated and the average measurement for that particular coordinate is generated. In Figures 3 and 4, we exhibited the representative coordinates of selected landmarks to illustrate the consistency between high, moderate, and low levels of coordinates measurements in different scenarios. The vertical axis of Bland–Altman plots shows the measurement variance of the selected landmarks, while the horizontal axis shows the average of the measurements. The zero line refers to the subject with zero measurement variance (highest reproducibility). The two dashed lines above and below the zero line indicate the subject with the highest variance between the two measurement sessions.

Table 2 shows the overall results of coordinates' reproducibility gained from intra- and inter-rater results of 160 volunteers. In addition to the overview, we display the results separately according to race and gender. Generally, reproducibility of most assessments was less than 1 mm (intra-examiner 87%, and inter-examiner 73.2%). In the Caucasian subgroups, the intra-examiner was 83.4% and inter-examiner was 71.1%; in Asian the intra-examiner was 79.7% and the inter-examiner was 72.5%. Among females, the intra-examiner was 79.7% and inter-examiner was 69.6%; among males, the intra-examiner was 82.6% and the inter-examiner was 73.2%. The highest reproducibility (<0.5 mm) coordinates were 45% (intra-rater) and 31.2% (inter-rater) of the 160 samples. The worst reproducibility (>1 mm) coordinates accounted for 13% (intra-rater) and 26.8% (inter-rater).

The error results of all landmarks on the x, y, and z axes after all samples were grouped by gender and race are shown in Tables 3 and 4, respectively (mean and SD). Additionally, we calculated the total error based on the results on each axis for each landmark. The landmarks were ranked from most reproducible to least reproducible for both intra- and inter-examiner assessments. Compared to the intra-rater assessments, we noticed that the corresponding inter-rater results showed poorer reproducibility.

Landmark accuracy for the Caucasian group ranged from 0.17 to 0.94 mm (intra-rater) and 0.20 to 1.38 mm (inter-rater) in the nose area, 0.44 to 0.61 mm (intra-rater) and 0.49 to 0.95 mm (inter-rater) in the eye area, 0.44 to 1.47 mm (intra-rater) and 0.52 to 1.75 mm (inter-rater) in the mouth area, 0.72 to 1.07 mm (intra-rater) and 1.19 to 1.63 mm (inter-rater) in the ear area, and 1.52 to 1.79 mm (intra-rater) and 1.23 to 2.04 mm (inter-rater) in the other areas.

Landmark accuracy for the Asian group ranged from 0.20 to 1.38 mm (intra-rater) and 0.25 to 1.13mm (inter-rater) in the nose area, 0.30 to 0.62 mm (intra-rater) and 0.54 to 0.85 mm (inter-rater) in the eye area, 0.47 to 1.52 mm (intra-rater) and 0.67 to 1.73 mm (inter-rater) in the mouth area, 0.86 to 0.97 mm (intra-rater) and 1.05 to 1.40 mm (inter-rater) in the ear area, and 1.23 to 2.04 mm (intra-rater) and 1.43 to 2.21 mm (inter-rater) in the other areas.

Differences in landmark reproducibility between Caucasian and Asian were concentrated in nose tip, alare, and nostril area, including Sn, Cc, Cm, Al right, Al left, Ac right, Nm of both sides, and Stb point. Several landmarks showed poor reproducibility in both Caucasians and Asians, namely Tri, Zy right, and Zy left. The landmarks with the most significant intra- and inter-group differences were Zy left with 0.8 mm in Caucasians and Pa right with 0.43 mm in Asians. Moreover, the measurement differences between intra- and inter-group assessment of landmarks in the Asian group were generally smaller than in the Caucasian group. The most and least reproducible landmarks in Asians were consistent with those in Caucasians (Table 3 & Supplement Table 1).

The accuracy of landmarks in the female subgroup ranged from 0.24 to 1.55 mm (intra-rater) and 0.32 to 1.88 mm (inter-rater) in the nose area, 0.31 to 0.53 mm (intra-rater) and 0.45 to 0.76 mm (inter-rater) in the eye area, 0.40 to 1.33 mm (intra-rater) and 0.53 to 1.83 mm (inter-rater) in the mouth area, 0.79 to 1.22 mm (intra-rater) and 1.40 to 1.99 mm (inter-rater) in the ear area, and 1.19 to 1.97 mm (intra-rater) and 1.17 to 2.25 mm (inter-rater) in the other areas.

The accuracy of landmarks in the male subgroup ranged from 0.21 to 1.45 mm (intra-rater) and 0.31 to 1.65 mm (inter-rater) in the nose area, 0.38 to 0.68 mm (intra-rater) and 0.49 to 0.85mm (inter-rater) in the eye area, 0.41 to 1.08 mm (intra-rater) and 0.50 to 1.57 mm (inter-rater) in the mouth area, 0.83 to 1.14 mm (intra-rater) and 1.20 to 2.06 mm (inter-rater) in the ear area, and 1.28 to 1.45 mm (intra-rater) and 1.03 to 2.43 mm (inter-rater) in the other areas.

Compared to females, landmarks concentrating on the nose and mouth areas had higher reproducibility in males in intra-rater, while landmarks in the eye area had poorer reproducibility in males. Moreover, the deviations between intra- and inter-rater in males were smaller than in females.

We did not notice significant differences in the ranking of landmark reproducibility between genders overall, except for the Sellion right and Nostril base point left and right. Among both female and male groups, Pronasale (prn) was the most reproducible landmark, while Zygion left was the least reproducible landmark (Table 4 and Supplement Table 1).

Some landmarks differed in the reproducibility levels in intra-rater and inter-rater assessments as follows: Cm and Stb were highly reproducible (<0.5 mm) in intra-rater and moderately (<1 mm) in inter-rater assessment for the Caucasian sample. In Asian sample, Se right and Se left were moderately reproducible (<0.5 mm) in intra-rater and poorly reproducible (>1 mm) in inter-rater assessment.

Bland–Altman plots are used to illustrate the consistency level between the values of each 3D coordinate (X, Y, and Z) for the facial landmarks. Some representative coordinates of facial landmarks are given in Fig. 3 to illustrate the high, moderate, and low levels of consistency between the measurements obtained from intra-rater assessment of the ethnic subgroup. Figure 3a indicates that the landmark Columella (Cm) was highly reproducible (<0.5 mm) in the X-plane for Caucasians and Asians. Figure 3b indicates that the landmark Menton (M) was moderately reproducible (>0.5 mm) in the Y-plane. Figure 3c indicates that the landmark Trichion (Tri) was poorly reproducible (>1 mm) in the Z-plane. Figure 4 exhibits some representative coordinates of the measurements obtained from the inter-rater assessment for facial landmarks in gender subgroup. Figure 4a indicates that the landmark Pronasale (Prn) was highly reproducible (<0.5 mm) in the X-plane for both females and males. Figure 4b indicates that the landmark Nasion (N) was moderately reproducible (>0.5 mm) in the Y-plane. Figure 4c indicates that the landmark Supratip break point (Stb) was poorly reproducible in the Z-plane (>1 mm).