Premature neonates' care can be improved by a quantitative evaluation of brain structures. In current clinical practice, first-line diagnosis is done qualitatively with 2D cranial ultrasonography (cUS) images. These patients would benefit from an evolution to a quantitative analysis based on 3D cUS. A way to obtain high quality 3D cUS images is to use a reconstruction algorithm based on the clinician acquisition motion. In this paper we assess the accuracy of such reconstruction by comparing brain structures between MRI and 3D renconstructed cUS. This comparison was performed based on manual contours, we obtained a mean Dice value of 0.72 ± 0.05 for thalami and ventricles and a mean volume error of 10 ± 5.7%. In addition, the contour interpolation method that we used significantly improved Dice and volume error compared to nearest neighbor approach.