A number of applications could benefit from an image processing algorithm that detects bone controus in ultrasound images. E.g.

• Scoliosis measurement. Bone segmentation could contribute to automatic bone landmark recognition. This would speed up the workflow and eventually eliminate most user interactions.
• Long bone fracture monitoring. Long bones are easier to recognize on ultrasound images, so this is a low hanging fruit. The goal is to measure the angle where the bone is broken. This would save X-ray cost and radiation, mainly in children.
• Spinal anesthesia navigation. There is no available automatic spine segmentation algorithm, especially for diseased spine. So spinal needle injections could only be navigated if at least part of the spine would be automatically recognized on ultrasound images, and would be displayed in the 3D navigation scene.

The following special characteristics of bone ultrasound images could be used for segmentation:

• The image is darker under the bone contour along the scan lines due to shadowing. Not completely black because of beam width and scatter.
• Bone surfaces are usually smooth. At a certain scale they can be approximated as flat surfaces. The bone probability could be increased at locations which are close to already detected bone planes.

By working on this project you can learn how to use the most popular medical image processing library (www.itk.org), and integrate your work in a large software toolkit used by many around the world for tracked ultrasound applications research.