Abstract
In recent years, there has been a growing interest to incorporate ultrasound into computer assisted orthopaedic surgery procedures in order to provide non-ionizing intra-operative imaging alternative to traditional fluoroscopy. However, identification of bone boundaries still continues to be a challenging process due low signal to noise ratio and imaging artifacts. The quality of the collected images also depends on the orientation of the ultrasound transducer with respect to the imaged bone surface. Shadow region is an important feature indicating the presence of a bone surface in the collected ultrasound data. In this work, we propose a framework for the enhancement of shadow regions from extended field of view spine ultrasound data. First bone surfaces are enhanced using a combination of local phase based image features. The combination of the phase features provides a more compact representation of vertebrae bone surfaces with supressed soft tissue interfaces. These enhanced features are used as an input to a L1 norm based regularisation method which emphasised uncertainty in the shadow regions. Validation on phantom and in vivo experiments achieve a mean dice coefficient value of 0.93 and 0.9 respectively.
