In order to overcome drawbacks of manually segmenting the maternal artifacts from 3D fetal ultrasound image even by experienced sonographers, an interactive segmentation approach was proposed based on level set framework. This algorithm could detect the boundary of objects by controlling the propagation speed function based on gradient orientation and scalar. In order to apply level set framework into 3D ultrasound segmentation, the Narrow-band scheme was put forward, which increases the segmenting speed and accuracy dramatically by restricting propagation process between two equidistant curves instead of updating it at all the points on the grid. Experiments were performed on 3D fetal ultrasound datasets from two different ultrasound machines containing 30 fetal images of different position, orientation and weeks. The result showed that this approach achieves fast and accurate segmentation of 3D fetal ultrasound image with size of 255×255×128 in 5 seconds on a dualcore 2.0 GHz computer.
