Detecting Stripe Artefacts in Ultrasound Parametric Images

Parametric ultrasound images visualising a patient’s brain perfusion often suffer from stripe artefacts – a darker, radially oriented image region originating from inhomogeneities in the cranial bone or the dura mater. We present an algorithm for detecting these artifacts with regard to automatic evaluation of parametric images. The algorithm exploits that stripe artefacts by virtue of their origin are always parallel to the ultrasound beam. After image preprocessing the algorithm amplifies vertical structures with a suitable filter kernel. It then continues to work on single possible stripe artefacts which are marked as such based on their average grey values in different subparts. Our results show that the algorithm recognizes stripe artefacts with an error rate of 0-7% in different images with artificial and real artefacts. Thus, it provides an important step towards automatic image evaluation since stripe artefacts are very close in their local appearance to pathological areas.