Cardiovascular diseases (CVDs) are the leading cause of disability and mortality in the world,
and their impact has been increasing in the past decades because of demographic trends such as population growth and ageing. Around 32% of all deaths are caused by CVDs, which mostly affiect the cardiac and cerebral vasculatures. Through the direct cost of the treatment of pathologies and the indirect money lost due to lack of patient productivity, vascular pathologies act as a large financial burden on the economy. Therefore, there is a huge demand for constant improvement of tools and methods for early diagnosis and effiective treatment of vascular pathologies.Due to the important role of imaging in diagnosis, planning and treatment of vascular pathologies, further improvements of these processes are immediately possible by advancing either the image acquisition or image analysis techniques.
One of the most typical cerebrovascular pathologies related to CVDs are aneurysms, baloonlike structures that bulge from a weakened portion of a vessel. Intracranial aneurysms have a prevalence from 1% to 5% of the world’s population and lead in the event of rupture to
stroke, a serious and life threatening condition. Although aneurysm rupture is a rather rare
event, ruptured aneurysms are the most common cause (85%) of nontraumatic subarachnoid
hemorrhages, which lead to stroke. To prevent such fatal events, either through preventive
measures or by surgical treatment, intracranial aneurysms need to be detected and assessed as
early as possible. In current clinical practice, a neuroradiologist detects and assesses aneurysms by visual inspection of a two-dimensional (2D) or three-dimensional (3D) angiographic image. Because angiographic acquisitions may diffier substantially in the level of contrast, resolution, noise and artifacts and because aneurysms are often surrounded by complex vascular networks and other structures, the detection based on visual inspection of angiographic images is clearly a difficult task. Moreover, to reliably detect all the aneurysms by interactive visual inspection of raw 3D images, even a trained neuroradiologist may require an excessive amount of time. Even more difficult is the quantitative assessment of vascular pathologies which typically is performed by measuring the aneurysm morphologic metrics in 2D angiographic images. While these measurements are rather simple to perform in 2D, the same metrics are essentially more accurate when measured in 3D, which, if done manually, is a much more challenging task. To improve the accuracy and reliability of detection and quantification of pathological structures in medical images, in general, and to reduce 3D image inspection and assessment times, substantial effiorts have been underway in the field of medical image analysis towards the development of tools for either automated or computer-aided pathology detection and quantification. While the aim of automated methods is a system completely independent of the neuroradiologist, the aim of computer-aided systems is to streamline tedious and time-consuming tasks so that the neuroradiologist is both effiective at image inspection and performs the detection and quantification accurately and reliably. This thesis concentrates on the development and validation of methods for computer-aided detection and quantification of intracranial saccular aneurysms in 3D angiographic images that are designed to assist a clinician towards a quicker and more reliable diagnosis and treatment of aneurysms. The main emphasis of the presented detection methods is the provision of high sensitivity and low specificity which is essential if to be used in clinical routine where an incorrect decision can adversely impact a patient’s life, whereas the design of the proposed quantification method aimed at producing accurate and robust morphologic measures that are unaffiected by aneurysms’ size and shape variations, and thus, providing a reliable mean for monitoring the state of the aneurysms through time.