查看文章

Methods: Based on MRI or CT, a 3D model of the liver, lesions, hepatic and portal vein was generated and matched with the intraoperative organ position either by 1) identifying common landmarks between the pre-and intraoperative image or 2) automatically using deformable modeling of hepatic and portal vasculature extracted from 3DUS using deep learning. Accuracy was assessed as Euclidean distance between the lesion imaged in the preoperative image and the one identified in intraoperative 3DUS.

Results: From March 2019 to May 2021, 45 navigated procedures were safely performed. Navigation provided additional localization features by overlaying preoperative 3D models of liver lesions, vasculature and biliary anatomy onto the intraoperative ultrasound. Accuracy, calculated in 36 patients, was 8.5±4.2 mm for the landmark method (n= 20), and 5.7±3.6 mm for automatic generation of the vasculature (n …