Investigation into variation and errors of a three-dimensional breast imaging system using multiple stereo cameras

Background: Errors derive from the variability seen in the repeat volume measurements for a particular individual. The aim of this study was to evaluate the variation and errors of a three-dimensional (3D) multiple stereo camera system for objective breast assessment. It was also investigated whether there was any correlation between these errors and the size of the breast.

Methods: A prototype eight camera multiple stereophotogrammetry system was utilized. The volumes of nine plaster breast models were determined by 3D imaging and the correlation between the size of the models and the variability of the measurements was investigated after ten repeats. The breasts of six live volunteers were examined following a specific protocol. The breasts were captured six times, three times each on two different occasions; from each breast capture a three-dimensional model was built and the breast volume was measured three times with breast analysis tool (BAT) software. This allowed for an assessment of the variability introduced at each stage of the measurement procedure. The correlation between the size of the breast and the variability of the measurements was investigated. Results of volume measurements by water displacement method, repeated ten times with the plaster models and six times with the live models, were used for comparison.

Results: The correlation between the size of the plaster models and the variability of the measurements revealed a significant correlation (p = 0.033), indicating that the larger the model, the more variable were the results. The correlation between the size of the breasts in the live models and the variability of the results revealed a non-significant correlation (p = 0.342), but there was a visible trend that the larger breasts showed more variable results. The average variation in the repeated measurements on each individual was found to be 32.95 cc across the two different poses, 19.43 cc across the various captures, and 28.32 across the different volume calculations with the BAT software. The error is less than what is clinically visible with the human eye, which is about 50 cc variation in volume by subjective assessment. As a proportion of the mean volume, namely 6.9%, 4.0% and 5.9% respectably, the values are small and indicate that the reproducibility of the system is good.

Conclusion: The 3D imaging system using multiple stereo cameras revealed a positive correlation between the size of plaster models and the breasts of live models and the reproducibility of the measurements, indicating that the variation was higher for the larger sizes. The relationship was significant in the plaster models but not in the live models, although a trend was observed. The assessment of the variation and errors of the system was part of the necessary procedure that should be considered for the validation of any new measurement technology for breast assessment.