The aim of this study is to assess the validity (accuracy) and reproducibility of images captured from a three-dimensional imaging system (3dMDface, Atlanta, GA), as well as the precision of measurements obtained from these images, using a symmetrical precision model.

Materials & Methods
Twenty images of a pre-calibrated symmetrical precision model (spherical ball) were captured - 10 in the morning and another 10 in the afternoon - 4 days apart using the 3dMDface imaging system. The last image for each morning/afternoon session was therefore taken 36 days after the first image was taken.

Validity of the system was assessed via 3 methods: 1) comparing the coordinate data sets of 13 landmarks on the model to the coordinate data sets obtained by a Coordinate Measuring Machine (gold standard) after matching by ordinary partial procrustes analysis (OPPA), 2) Euclidean Distance Matrix Analysis (EDMA) to test for difference in form produced by the two systems, and 3) symmetry of the symmetrical spherical ball (left and right sides, as well as top and bottom halves) were assessed on each image, as an indicator of system accuracy and to qualitatively locate any distortions.

Reproducibility error of the images (capture error) was calculated as the surface-to-surface root mean square distance between repeat images of the same object after computer registration. The effect of time of image capture (morning or afternoon), as well as the effect of time lag (up til 36 days) on the accuracy and reproducibility of the system was investigated using one-sample t-tests and univariate ANOVA.

Operator error (precision of landmark localization) was calculated as the distance between corresponding landmarks of repeated digitizations of the same image. Precision of various linear measurements, area and volume measurements were also calculated.

Results showed that 3dMDface system is highly accurate to within ± 0.117mm, producing images that are the same in form as the "true" object, represented by a high-precision coordinate measuring machine (CMM). In addition, there were distinct differences between the various methods employed to assess accuracy: EDMA was the most sensitive to differences in form but lacks easily interpretable graphic output and is limited to the number of distances evaluated. Computer registration of the 2 mirror halves of the symmetrical ball may be less sensitive, but is able to produce a contour map of the surface difference, and allows thousands of point data on the object image to be assessed. OPPA was the least sensitive test.

Images obtained via 3dMDface are highly reproducible, with a RMS surface distance difference of 0.089mm ± 0.0420mm between repeat images. The time of day in which the images were captured (morning versus late afternoon), had no effect on accuracy of the images, but had a small statistically significant effect on reproducibility. The RMS difference of 0.043-0.049mm was however deemed clinically insignificant.

Landmarks on 3D images produced by 3dMD can be accurately located to within 0.06mm for the y and z axis, and to within 0.03mm for the x-axis. Bland-Altman analysis and other estimates of precision showed high precision for linear distance, surface area, and volume measurements.

Finally, great variability exists in the methods used to assess reliability of 3D capturing devices. It is hoped that development of the use of an exactly symmetrical precision ball in this study may form part of a standard protocol for reliability testing of 3D imaging systems in future.