Imaging bone with ultrasound has become a popular trend in research, since it uses no radiation compared to X-ray based methods. For many clinical tasks (including orthopaedic surgery and diagnosing developmental dysplasia of the hip), it is also important to localize where the bone is in the US image. This is known as bone segmentation.
Over the last decade or so, several (semi-)automatic methods of segmenting bone in ultrasound have been published. Despite this, there is no common dataset on which these methods can be benchmarked. Since research is performed on closed datasets, it is difficult to directly compare methods and their performances with each other. This usually results in many methods being designed in parallel with similar performances, instead of methods being built/improved on each other. Many other computer vision fields or medical imaging fields have benefited from a common dataset and a way to systematically evaluate segmentation techniques. This is exactly what we are trying to instantiate for ultrasound bone segmentation!
Feel free to reach out to us if you're at CAOS - or send me an email by clicking the link below!