Fig. 4.

Segmentation of ultrasound volumes acquired before resection. In each example, the axial, sagittal and coronal views are shown in the first, second and third row, respectively. In the first column, the original ultrasound volume is exhibited, in the second column, the manual annotation performed on the axial view and projected in the other two views is shown, in the third column, the segmentation result obtained by the 3D U-net for the same volume of interest is displayed. In each example, a pointer (intersection of yellow crossing lines) highlights the same volume position in the three views. Our method correctly segments the main structures. Moreover, structures wrongly not included in the manual annotations are correctly detected by the trained neural network (purple squares). However, in image d, pathological tissue correctly excluded in the original masks is wrongly segmented by our method (blue squares in axial view)