PhD Defense: Automated Detection and Classification of Intracranial Aneurysms based on 3D Surface Analysis

Abstract

Intracranial aneurysms are localized, abnormal arterial dilations with a variable risk of rupture, which can lead to subarachnoid hemorrhage (SAH), a condition associated with high morbidity and mortality. The majority of non-traumatic SAH cases are caused by ruptured intracranial aneurysms. Accurate detection can decrease a significant proportion of misdiagnosed cases. However, only a small percent of all detected incidental aneurysms proceed to rupture and preventive treatment carries risks of complications, thus creating a need for aneurysm rupture risk stratification tools to help guide the treatment of these asymptomatic lesions. This research investigates the two major areas of intracranial aneurysm analysis - aneurysm detection and rupture risk classification. First a method for automatic detection of intracranial aneurysms is proposed. Applied to the segmented cerebral vasculature, the method detects aneurysms as suspect regions on the vascular tree, and is designed to assist diagnosticians with their interpretations. The method is imaging modality independent and was tested on magnetic resonance imaging, computed tomography and 3D angiography data. Second, a new approach to morphological analysis of the 3D shape of an aneurysm is presented as a differentiator of rupture risk status in cerebral aneurysms. The writhe number is introduced as a novel surface descriptor which proves useful in both detection and classification studies. In addition to the writhe number, 3D shape descriptors based on surface curvature and centroid-radii model are proposed and investigated for rupture risk classification. The combined use of all these new features yields very promising results for predicting rupture risk in intracranial aneurysms. In experiments, the aneurysm detection method achieved 100% sensitivity, independent of modality. Depending on the imaging modality, there are between 0.66 and 5.7 false positives per study; the worst case performance is comparable to that of existing detection research. The classification method resulted in a ~20% increase in prediction accuracy, compared to other commonly used shape indexes. These results support the utility of writhe number aneurysm shape analysis as a high order descriptor with potential clinical use in intracranial aneurysm detection and rupture risk stratification.