The diagnosis and prognosis of brain anomalies in preterm neonates can be improved by the quantification of the cerebral volumes of different brain areas, such as the ventricles [1]. In the current clinical routine, a diagnosis is made based on 2D images of the brain, without making quantitative assessment of the volumes of the cerebrospinal fluid in the ventricles. We propose a semi-automated segmentation method to determine the volume of the ventricles, the choroid plexus, the cavum vergae and the cavum pellucidum in 3D ultrasound images of the brain. The algorithm requires only three seed points given by the operator. We show the good agreement between the segmented contours and manually segmented images on clinical data acquired with a commercial neonatal head probe.