Abstract
A challenging problem in ultrasound based orthopaedic surgery is the identification and interpretation of bone surfaces. Recently we have proposed a new fully automatic ultrasound bone surface enhancement filter in the context of spine interventions. The method is based on the use of a Gradient Energy Tensor filter to construct a new feature enhancement metric, which we call the Local Phase Tensor.
The goal of this study is to provide further improvements to the proposed filtering method by incorporating a-priori knowledge about the physics of ultrasound imaging and salient grouping of enhanced bone features.
Typical ultrasound scan of the spine, there is a large soft tissue interface present close to the transducer surface with high intensity values similar to those of the bone anatomy response. Typical ultrasound image segmentation or enhancement methods will be affected by this thick soft tissue response. In order to weaken this soft tissue interface we calculate a new transmission map where features deeper in the ultrasound image have higher transmission values and shallow features have lower transmission values. The calculation of this new US transmission/attenuation map allows the proposed image enhancement method to mask out erroneous regions, such as the soft tissue interface, and improve the accuracy and robustness of the spine surface enhancement. The masked US images were used as an input to the LPT image enhancement method. In order to provide a more compact spine surface representation and further reduce the typical US imaging artifacts and soft tissue interfaces we calculate saliency Local Phase Tensor features. The saliency images are computed using Difference of Gaussian filters.
Qualitative results, obtained from in vivo clinical scans, show a strong correspondence between enhanced features and the actual bone surfaces present in the ultrasound scans. Future work will include the extension of the proposed method to 3D and validation of the method in the context of intra-operative ultrasound image registration in CAOS applications.
