Physics-based Virtual Environment for Training in Vascular Interventional Radiological Procedures
Yi Song, Ken Brodlie, Andy Bulpitt and David Kessel
This work is part of a EPSRC funded CRaiVE project to develop a physics based Virtual Environment (VE) for training in vascular interventional radiological procedures. It aims to provide an alternative to apprenticeship training using physical and animal models. One of the major tasks in this project is to create a variable range of 3D geometry of anatomy, which requires fast and precise segmentation of varied abdominal structures, e.g. liver, kidney and blood vessels etc. Particular interests are in those cases where typical pathology is presented. The data in this study therefore comes from patients with various pathologies and is obtained from different sources using different protocols which vary in quality and resolution. These diversities increase the variability of the organs in both shape and texture.
We are developing a general solution to fast segmentation of abnormal structures with minimal user interaction. In contrast to previous work focusing on either specific image type or specific structures, our segmentation system is able to cope with a wide variety of structure shapes and image modalities. An example result of MRA image segmentation can be found in Gallery. The process of segmenting blood vessels from a CT image is shown in Figure 1.
Figure 1 Blood vessel segmentation from CTA image.
Having the blood vessels segmented, we can automatically detect the calcifications on the vessel wall, as show in Figure 2.
Figure 2 Calcification detection from CT image in 2D sagittal view and its 3D reconstruction. Left: original grey image. Middle: artery and calcification segmentation results are mapped onto the original image. Right: 3D reconstruction.
As metioned above, although this project mainly focuses on the study of blood vessels, our work is not limited to vascular structure segmentation. In Figure 3, we demonstrate a liver segmentation where tumors and hepatic (portal) veins are also presented.
Figure 3 Liver segmentation (right) from constrast-enhanced CT image (left).
Often only the diseased organ is highlighted in patient CT scans. However, the VE simulating the interventional radiological procedures requires neighbouring anatomical features. Therefore, in Figure 4, we demonstrate that our approach can be adapted to the task of segmenting those non-pathological structures which often have similar intensity value to their surrounding tissues, since they normally are not contrast-enhanced in the CT images.
Figure 4 Create virtual human (right) from the CT image (left).