Validation and reproducibility of a high-resolution three-dimensional facial imaging system

https://doi.org/10.1016/j.bjoms.2007.04.017Get rights and content

Abstract

Objective

To assess the accuracy and reproducibility of a high-resolution three-dimensional imaging system (Di3D).

Design

The three-dimensional imaging system was validated in vitro using 12 adult facial plaster casts, which had landmarks marked, and the positions of the landmarks on the three-dimensional images captured by Di3D were compared with those obtained by a coordinate measuring machine (CMM).

Methods

Operator error was measured by repeatedly locating landmarks on the three-dimensional image. Reproducibility error of the images was calculated by capturing three-dimensional images of the facial casts on two separate occasions; the Euclidean distance between the two matched sets of coordinates was then calculated. The Di3D system error was assessed by calculating the three-dimensional global positions of landmarks on the three-dimensional images and comparing them with those obtained by CMM (gold standard).

Results

The operator error in placement of landmarks on the three-dimensional model was 0.07 mm, range 0.02–0.11 mm. The reproducibility of the Di3D capture was 0.13 mm, range 0.11–0.14 mm. The mean distance between the CMM and Di3D landmarks, which constitutes the Di3D system error, was an average of 0.21 mm, range 0.14–0.32 mm.

Conclusions

The Di3D system error was within 0.2 mm, which is clinically acceptable, and offers considerable improvement in stereophotogrammetry for facial capture and analysis.

Introduction

Accurate and valid three-dimensional measurements of the surface anatomy of facial soft tissues are fundamental for the objective analysis of craniofacial deformities. Many techniques for taking three-dimensional measurements of soft tissues are available, including biostereometrics,1 morphanalysis,2 laser scanning,3 three-dimensional digitisation,4 Moiré scanning,5 and ultrasonography.6 Each technique has potential disadvantages: for example, laser scanning takes approximately 15 s to complete a full facial scan, and morphanalysis uses equipment that is extremely elaborate and expensive, and is time-consuming.

Recording the soft tissue's outline with stereophotogrammetry is probably the most promising option.7 The validity and accuracy of a low-resolution stereophotogrammetry three-dimensional imaging system (C3D) that is used to capture facial images has been reported previously.8 The process relied on projection of texture to provide sufficient information in the images to match the two sides of the face and construct the three-dimensional facial model accurately. The study concluded that operator error was within 0.2 mm of the true coordinates of the landmarks, and that the C3D system was accurate to within 0.4 mm.

Projection of texture prolonged the three-dimensional capture of the subject by up to 10 ms; the method is not standard, and required complex projection machinery. Projection of texture also limited the achievable working volume of a system, because the projectors used in the system required wide apertures for optical efficiency, but this resulted in a small projection depth of field. By using commercially available professional high-resolution colour digital cameras (4000 pixels × 3500 pixels) it is now possible to capture images that resolve local details of linear densities approaching 0.1 mm/pixel on human faces. At this resolution there is enough information about local texture to achieve reliable area-based stereo matching. Projection of texture is therefore no longer required to achieve a good three-dimensional reconstruction of the human face based on high-resolution stereo-pair images of the subject and the capture time is shortened to 1 ms.

The aim of this study was to assess the accuracy and reproducibility of the high-resolution three-dimensional imaging system (Di3D, Dimensional Imaging, Hillington Park, Glasgow, UK) in locating landmarks on adult, full-face plaster casts compared with measurements taken using a coordinate measuring machine (CMM) based on a touch probe of known accuracy (Aberlink Ltd., Gloucestershire, UK). The geometric accuracy of the CMM as measured by the manufacturer was 0.01 mm. Maximum touch probe accuracy was 0.001 mm.

Section snippets

Material and methods

Twelve healthy adult volunteers were recruited for the study, six men and six women. A full facial alginate impression (Xantalgin Select fast set, Heraeus Kulzer GmbH, Germany) was taken for each subject, and a stone model was constructed (Kaffir D, East Sussex, UK) (Fig. 1). A hexagonal locking plate (Manfrotto, Italy) was secured to the base of each model. Ten points were marked on each facial cast using a fine dental probe and the indent was highlighted with black ink (Lumocolor, Staedtler,

Variation in CMM measurement

The variation in CMM measurements is shown in Table 1, and given as a scattergram in Fig. 4.

Operator error

The magnitude of error associated with placement of landmarks using the Di3D system is shown in Table 2 and the results are given as a scattergram in Fig. 5.

Reproducibility of the Di3D system

The magnitude of error associated with capture and digitising landmarks of the three-dimensional images on two separate occasions, is shown in Table 3. It is important to emphasise that the reported errors also include the operator error of an

Discussion

The validity of the three-dimensional stereophotogrammetry imaging system (Di3D) was compared with that of a well-accepted highly accurate three-dimensional coordinate measuring machine using facial plaster models.

The operator error was low, and was within 0.07 mm of the true coordinates of the landmarks obtained by CMM. The variation in identification of landmarks for both systems was virtually identical (0.1 mm). This is not surprising, as the landmarks that were marked on the casts and

Conclusion

We conclude that the Di3D system error was within 0.2 mm, which is clinically acceptable, and is a major improvement in the use of stereophotogrammetry for facial capture and analysis.

References (15)

There are more references available in the full text version of this article.

Cited by (97)

  • Asymmetry index for the photogrammetric assessment of facial asymmetry

    2022, American Journal of Orthodontics and Dentofacial Orthopedics
View all citing articles on Scopus
View full text