Engineering a 3D ultrasound system for image-guided vascular modelling

Abstract

Atherosclerosis is often diagnosed using an ultrasound (US) examination in the carotid and femoral arteries and the abdominal aorta. A decision to operate requires two measures of disease severity: the degree of stenosis measured using B-mode US; and the blood flow patterns in the artery measured using spectral Doppler US. However other biomechanical factors such as wall shear stress (WSS) and areas of flow recirculation are also important in disease development and rupture. These are estimated using an image-guided modelling approach, where a three-dimensional computational mesh of the artery is simulated. To generate a patient-specific arterial 3D computational mesh, a 3D ultrasound (3DUS) system was developed. This system uses a standard clinical US scanner with an optical position sensor to measure the position of the transducer; a video capture card to record video images from the scanner; and a PC running Stradwin software to reconstruct 3DUS data. The system was characterised using an industry-standard set of calibration phantoms, giving a reconstruction accuracy of ± 0.17 mm with a 12MHz linear array transducer. Artery movements from pulsatile flow were reduced using a retrospective gating technique. The effect of pressure applied to the transducer moving and deforming the artery was reduced using an image-based rigid registration technique. The artery lumen found on each 3DUS image was segmented using a semi-automatic segmentation technique known as ShIRT (the Sheffield Image Registration Toolkit). Arterial scans from healthy volunteers and patients with diagnosed arterial disease were segmented using the technique. The accuracy of the semi-automatic technique was assessed by comparing it to manual segmentation of each artery using a set of segmentation metrics. The mean accuracy of the semi-automatic technique ranged from 85% to 99% and depended on the quality of the images and the complexity of the shape of the lumen. Patient-specific 3D computational artery meshes were created using ShIRT. An idealised mesh was created using key features of the segmented 3DUS scan. This was registered and deformed to the rest of the segmented dataset, producing a mesh that represents the shape of the artery. Meshes created using ShIRT were compared to meshes created using the Rhino solid modelling package. ShIRT produced smoother meshes; Rhino reproduced the shape of arterial disease more accurately. The use of 3DUS with image-guided modelling has the potential to be an effective tool in the diagnosis of atherosclerosis. Simulations using these data reflect in vivo studies of wall shear stress and recirculation in diseased arteries and are comparable with results in the literature created using MRI and other 3DUS systems.