Intra-operative ultrasound is an increasingly important imaging modality in\nneurosurgery. However, manual interaction with imaging data during the\nprocedures, for example to select landmarks or perform segmentation, is\ndifficult and can be time consuming. Yet, as registration to other imaging\nmodalities is required in most cases, some annotation is necessary. We propose\na segmentation method based on DeepVNet and specifically evaluate the\nintegration of pre-training with simulated ultrasound sweeps to improve\nautomatic segmentation and enable a fully automatic initialization of\nregistration. In this view, we show that despite training on coarse and\nincomplete semi-automatic annotations, our approach is able to capture the\ndesired superficial structures such as \\textit{sulci}, the \\textit{cerebellar\ntentorium}, and the \\textit{falx cerebri}. We perform a five-fold\ncross-validation on the publicly available RESECT dataset. Trained on the\ndataset alone, we report a Dice and Jaccard coefficient of $0.45 \\pm 0.09$ and\n$0.30 \\pm 0.07$ respectively, as well as an average distance of $0.78 \\pm\n0.36~mm$. With the suggested pre-training, we computed a Dice and Jaccard\ncoefficient of $0.47 \\pm 0.10$ and $0.31 \\pm 0.08$, and an average distance of\n$0.71 \\pm 0.38~mm$. The qualitative evaluation suggest that with pre-training\nthe network can learn to generalize better and provide refined and more\ncomplete segmentations in comparison to incomplete annotations provided as\ninput.\n