Aneurysm Rupture Risk Analysis and Risk Prediction Modeling Based on CFD Simulations and Statistical Learning



Journal Title

Journal ISSN

Volume Title



Cerebral aneurysms are a common vascular disease occurring in about 2-3% of the general population. While most aneurysms remain asymptomatic and never rupture during a patient's lifetime, aneurysm rupture leads to subarachnoid hemorrhage, a subtype of stroke, which is associated with high mortality, morbidity, and substantial economic burden. Today, an increasing number of unruptured aneurysms are diagnosed as incidental findings. Since the risks related to treatment and complications outweigh the relatively low natural aneurysm rupture risk, about 1% per patient per year on average, the assessment of a patient's individual aneurysm rupture risk is essential. The mechanisms leading to aneurysm rupture are not fully understood, complicating the risk assessment. Different risk factors have been associated with aneurysm rupture in previous studies, including hemodynamic, morphological, anatomical, and patient-related parameters. Combining such factors into a statistical model for predicting aneurysm rupture could assist physicians in their treatment decision of unruptured aneurysms. Currently available models either do not include hemodynamic or morphological information, or are based on small sample sizes. The aim of this dissertation therefore was the development of a statistical model for aneurysm rupture combining the different types of risk factors and using (cross-sectional) data of large patient cohorts with about 2,000 aneurysms for model development and evaluation. For this purpose, hemodynamic information for each of the cases was obtained by patient-specific computational fluid dynamics (CFD) simulations. Hemodynamic parameters were particularly included in the analyses and models because of their role in aneurysm development, growth, and rupture through biomechanical signaling mechanisms in the vessel wall. Towards the aim described above, first, based on a univariate analysis it was found that certain hemodynamic and morphological factors were associated with aneurysm rupture. Motivated by this fact, a statistical model was trained, which achieved a good predictive performance in the training and external validation data. Aneurysms with high rupture probabilities were characterized by stronger and more complex flows as well as less regular shape. Furthermore, aneurysm location in the cerebral vasculature was particularly important for prediction of aneurysm rupture status. Related to this, it could be shown that a model trained specifically for aneurysms at the posterior communicating artery - a common site of aneurysms - could improve the predictions for aneurysms at this location. The general statistical model was then further extended to other populations by adding data from Japanese and Finnish patients to the previous US training and European (not Finnish) external validation data and taking population-specific associations with aneurysm rupture into account. Finally, the influence of variations of inflow boundary conditions for the CFD simulations on the computed hemodynamics and trained statistical models was assessed to address the inter- and intra-patient variation of blood flow in the cerebral arteries, and approaches for incorporating these variations into the statistical models were suggested. Overall, it was shown that the combination of hemodynamic, morphological, anatomical, and patient-related factors in a statistical model enabled accurate prediction of aneurysm rupture status. Once evaluated with longitudinal data, translation of such a model into the clinic could support physicians in their treatment decisions of unruptured aneurysms and potentially improve patient outcome.