Application of image-based computational modeling to vascular access surgery

Abstract

An arteriovenous fistula is a connection between an artery and vein, which is typically created in the arm, and used as vascular access in hemodialysis (i.e. blood filtration) of end-stage renal disease patients. The surgical creation of an arteriovenous fistula is hampered by short-term and long-term complications, which occur in 20%–50% of all cases. Most of these complications are flow related: too low flows can result in inefficient dialysis and thrombus formation, too high flows can result in lower arm ischemia (i.e., shortage of blood supply) and cardiac failure. The flow after surgery depends on the surgical procedure that is taken, which is patient-specific, therefore an individual planning is required. To gain insight into the complex effect of vascular access creation on the blood circulation, an image-based computer model of the circulation is developed. This might aid the surgeon during operation planning when additional preoperative information of the patient hemodynamics is supplied, and when surgical outcome is predicted. The general process for imagebased computational modeling consists of: 1) Image acquisition; 2) Image processing; 3) Computer model calculations; 4) Results visualization; 5) Interpretation. The aim of thesis was to develop and validate an image-based computational modeling framework for the application of vascular access surgery. For this application, image acquisition was performed by magnetic resonance (MR) imaging of the arm, image processing was used to determine blood vessel diameters, computer calculations were used to compute the flow of several fistula configurations, and visualization was used to present the predicted flows to the surgeon for interpretation. As a chain is only as strong as its weakest link, the sub processes were evaluated, starting with an evaluation of the accuracy and precision of automatic diameter measurements (image processing) in MR images, using a realistic vascular phantom. To set our method against clinical practice, the results were compared with the results of 15 experts, which used manual diameter measurement methods. This study concluded that the automatic method improves both accuracy and precision, and it was therefore recommended to apply an automatic diameter measurement method for quantitative vascular diameter measurements. Next, the image-based modeling framework was developed and validated against preoperative blood flow measurements for 8 healthy volunteers and 8 end-stage renal disease patients. The results demonstrated that with the image-based modeling framework an accurate description of the patient’s preoperative hemodynamic state was obtained. An interactive visualization of the geometric model and the hemodynamic results provided the vascular surgeon with additional information in the preoperative planning stage. To replace the current contrast-enhanced magnetic resonance angiography (CEMRA) image acquisition protocol with a non-invasive alternative, a non contrastenhanced MRA (NCE-MRA) acquistion protocol was developed and feasibility was demonstrated. Overall, the NCE-MRA are of diagnostic quality and venous depiction is superior for NCE-MRA, due to its high vessel-to-background contrast. The relation between NCE-MRA-based diameter measurements and ultrasoundbased diameter measurements was investigated, because surgical decision making currently relies on ultrasound. For 27 end-stage renal disease patients, diameter measurements were conducted on standardized locations using both modalities and compared. This revealed substantial differences between NCE-MRA and ultrasound, indicating that extreme caution should be exercised when replacing one diameter measurement modality by an other. As a final validation of the assimilated image-based modeling framework, the model’s predictions were compared with postoperative measurements. The results demonstrated that predictions overlap with postoperative flows for most of the patients. For the other patients, the differences can be significantly decreased by taking into account geometric details, which are noticeable in the NCE-MRA data. In conclusion, an image-based modeling framework was successfully developed and applied to vascular access surgery for prediction of postoperative arm inflow. All the intermediate process steps have been developed, (quantitatively) analyzed, and integrated to a functioning platform for assisting the vascular surgeon. This thesis successfully demonstrated the potential of this framework for patient-specific prediction of surgical outcome. At this time the image-based modeling platform is ready for validation in a multi-center study with a large population of end-stage renal disease patients, and is ready for exploration of other surgical domains.