Automation of the preoperative image processing steps for ultrasound based navigation

Ultrasound based navigation is a flexible, fast, and robust method for intraoperative navigation, where intraoperative 3D ultrasound is used for the registration procedure. To establish ultrasound based navigation in the clinical routine, it is necessary to automate the preoperative image processing steps. The task of this process is the extraction of the bone surface from the preoperative CT or MRI data. For the automation we developed an image processing pipeline. We designed a model of an ultrasound scan, which consists of a scan path and scan properties as transducer shape and width. The scan path was attached to anatomical landmarks. Based on these landmarks the model was registered within the preoperative image data and an ultrasound scan was simulated in the data to extract the bone surface visualised in ultrasound images. Additionally for complex structures as the lumbar spine it is necessary to separate single vertebrae. This segmentation was done by a shape-based level set method. The segmentation result was combined with the extracted bone surface, to assign the correct surface points to the vertebra. The ultrasound registration with the described surface extraction method was evaluated by applying the proposed procedure on phantom and patient data. To estimate the overall accuracy, phantoms of the lumbar spine and the femur were used to compare the ultrasound registration with an accurate point-based registration. Therefore, 100 ultrasound registrations were compared with the reference registration and target registration errors were calculated for different anatomical regions. For instance, at the phantom of the femur the mean RMS error for all targets was 0.74 mm, where 0.64 mm was the systematic and 0.36 mm was the statistical error. The results lie within an admissible range for intraoperative navigation.