Studies on the assessment of abdominal aortic aneurysm rupture risk

• Main Author: Canchi, Tejas
• Other Authors: Ng Yin Kwee, Eddie
• Format: Theses and Dissertations
• Language: English
• Published: 2018
• Subjects: DRNTU::Engineering::Bioengineering
• Online Access: http://hdl.handle.net/10356/73349
• Institution: Nanyang Technological University

Abdominal Aortic Aneurysms (AAA) are a type of cardiovascular disease that affect the infra renal aorta, where the normal aorta dilates to more than two times its original diameter. Progressive dilation leads to weakening of the aortic wall and eventually, rupture. Pathophysiological, biomechanical and genetic factors contribute to the genesis, growth and rupture of the AAA. Current clinical management of AAA is based on the maximum transverse diameter of the dilated lumen to recommend surgical intervention. However, several additional factors play a role in the progression to rupture. Therefore, a patient specific assessment based on biomechanical factors and antecedents can lead to the development of a predictive tool that can be implemented at the bedside. This thesis has investigated a biomechanical as well as machine learning/statistical approach to developing an accurate patient specific predictive tool for rupture assessment in Asian AAA patients. This was deemed necessary due to the differences in morphology between Asian and Caucasian patients. These differences have caused difficulties with the graft delivery during Endovascular Aneurysm Repair (EVAR) and subsequent follow up of AAA Asian patients. The clinical significance of these differences can be brought out through a comparison of the geometric parameters in each cohort along with the mechanical stresses developed in these aneurysms. A comparison of 19 patients each from the Asian and Caucasian patient groups was conducted to establish the differences in morphology and hence, the biomechanics of rupture in each cohort. This was done using a combination of in-house segmentation codes and commercial finite element analysis (FEA) software. Subsequently, statistical analysis was also applied to generate geometric surrogates to predict the wall stresses in the patient groups. The study resulted in significant differences in peak wall stress among Asian patients with high maximum diameters when modeled with non-linear constitutive material models. The means of the biomechanical stresses between the two patient groups were not significant between the two patient groups. It was seen that the geometric indices that were significantly correlated to the spatially averaged wall stress in the Asian patient group were maximum diameter, proximal neck diameter, L2 norm of the mean curvature (MLN), square root sum of the Gaussian curvature (KM), and the L2 norm of the minor principal curvature (K2LN) while in the Caucasian patient group it was only distal neck diameter. The resulting six geometric indices can be used as geometric surrogates to quantify wall stresses in the patients. The five indices in addition to maximum diameter can predict the spatially averaged wall stresses with an accuracy of 72.87% in the Asian patient group and 74.74% in the Caucasian patient group. Machine learning algorithms (MLA) as a possible approach to hasten the diagnostic process was explored using multiple classifiers such as decision trees, support vector machines (SVM), Naïve Bayes and logistic regression. The most suitable classifier analyzing a cohort of 312 patients (155 AAA and 157 normal) with geometric and patient antecedent attributes, was determined as the Naïve Bayes algorithm. Further, a feature selection algorithm was applied to obtain the attributes that are significantly correlated to rupture. These included the proximal neck diameter, the neck length and the right iliac artery diameter in addition to the maximum diameter of the AAA. A further investigation involving 155 patients’ data was done to estimate whether only antecedent attributes can accurately predict rupture in the cohort. Using only attributes based on patient history, it was observed that the MLA can accurately extract the most significant clinical features that are correlated to AAA rupture. It was determined that machine learning algorithms can be used in synergy with the biomechanics to estimate rupture risk in the cohort. A fluid structure interaction (FSI) based model is developed on patient specific geometry data originating in Singapore to establish the biomechanics of rupture in Asian AAA. A comparative analysis of two patient models with differing maximum diameters and literature based inlet and outlet boundary conditions was carried out by extracting biomechanical parameters such as principal stresses and wall shear stresses from the simulation. The patient model with the larger maximum transverse diameter developed lower principal and wall shear stresses than the one with a smaller value of the diameter. In order to validate computational results, experimental methods need to be developed to accurately construct a physical model of the system being investigated. A patient specific silicone AAA phantom was used in a benchtop flow loop to mimic the fluid structure interaction between the aortic wall and blood. Realistic boundary conditions were imposed using a physical Windkessel model that mimics the impedance in the abdominal aortic system. Resistance and capacitance modules were built to establish the impedance values that generate realistic pressure values at the outlet. Future work will involve analysis of a larger cohort of patients to compare the Asian and Caucasian cohorts, to develop a more robust machine learning model that can predict rupture and incorporation of patient specific boundary conditions in FSI simulations. Experiments using in vivo impedances will also be done to establish realistic boundary conditions in the AAA.