Birth-weight prediction by two- and three-dimensional ultrasound imaging

Objectives To compare the accuracies of birth-weight predicting models derived from two-dimensional (2D) ultrasound parameters and from total fetal thigh volumes measured by three-dimensional (3D) ultrasound imaging; and to compare the performances of these formulae with those of previously published equations. Methods A total of 210 patients were evaluated to create a formula-generating group (n = 150) and a prospective-validation group (n = 60). Polynomial regression analysis was performed on the first group to generate one equation based on 2D ultrasound measurements, one based on fetal thigh volume measured by the multiplanar technique (ThiM) and one based on fetal thigh volume obtained by the Virtual Organ Computer-aided AnaLysis (VOCAL (TM)) method (ThiV). Paired-samples t-tests with Bonferroni adjustments were used to compare the performances of these equations in the formula-finding and the prospective-validation groups. The same approach was used to compare the accuracies of the new 2D and 3D formulae with those of both original and modified 2D equations from previous publications, as well as the 3D model reported by Chang et al. Results The formulae with the best fit for the prediction of birth weight were: estimated fetal weight (EFW) = -562.824 + 11.962 x AC x FDL + 0.009 x BPD2 x AC(2) (where AC is abdominal circumference, FDL is femur diaphysis length and BPD is biparietal diameter), EFW = 1033.286 + 12.733 x ThiM, and EFW = 1025.383 + 12.775 x ThiV. For both the formula-generating and the prospective-validation groups, there were no significant differences between the accuracies of the new 2D and 3D models in the prediction of birth weight. When applied to our population, the performances of the modified and original versions of the previously published 2D equations and the performance of the original 3D formula reported by Chang et al. were all significantly worse than our models. Conclusions We believe that the greatest sources of discrepancy in estimation of birth weight are the phenotypic differences among patients used to create each of the formulae mentioned in this study. Our data reinforce the need for customized birth-weight prediction formulae, regardless of whether 2D or 3D measurements are employed. Copyright (C) 2009 ISUOG. Published by John Wiley & Sons, Ltd.