查看文章

The liver is a common location for primary cancer and metastatic disease. Currently, ultrasound is the only imaging modality that is widely accepted and integrated into a surgical workflow, automatic registration with preoperative imaging would provide great value in determining a resection plan. An initial registration is performed based on the ultrasound probe orientation and one point translation. The centerline of automatically, using a 3D U-Net, segmented intraoperative ultrasound is registered with the preoperative vasculature model. In visually successful registrations we acquire a target registration error of 12.29 ( +/- 4.93 mm), however, 55 % of the registrations fail expectantly due to a relatively big volume difference with respect to the ultrasound information that is acquired. Manually adjusting these cropped volumes reduces the TREs over all volumes from 47.32 (+/- 25.71 mm) to 25.66 (+/- 10.48 mm). In conclusion, we demonstrate a fast (69.74 +/- 14.6 seconds) deep learning based hepatic vasculature registration pipeline. Given that the ultrasound acquisitions do not contain the vena cava or gallbladder, and span a large part of the hepatic vasculature, our approach looks promising. Further optimization of automatically acquiring similar point clouds is expected to stimulate the adaptation of surgical navigation.