The brain at risk: Detection and quantification of cerebrovascular disease

Abstract

Cerebrovascular disease results in morphological and functional alterations in the intracranial macro- and microvasculature. Early detection of such alterations is key for better understanding of cerebrovascular disease mechanisms, and in taking preventative measures against cerebrovascular disease development. In this thesis, we aimed to quantify alterations in the vessel walls, brain parenchyma, and blood flow using both high-resolution post-mortem 7T magnetic resonance imaging (MRI), histopathological analysis, and in vivo MRI in volunteers, and in patients with a history of cerebrovascular disease.

High-resolution post-mortem 7T MRI and histopathological analysis of circle of Willis specimens was performed in Chapter 2, where we investigated arterial remodeling caused by hypertension. We observed no significant differences in vessel wall area and thickness, between hypertensive patients and normotensive controls. Histological analysis showed early and advanced atherosclerotic plaques for almost all vessels of the circle of Willis for both groups. The collagen to smooth muscle ratio, as a measure for arterial stiffness, was significantly higher for the internal carotid artery in the hypertensive group.

In Chapter 3, we investigated the limitations of vessel wall MRI regarding the accuracy of vessel wall thickness measurements. Using high-field 7T MRI of post-mortem circle of Willis specimens, we have shown that with conventional measurement methods accurate thickness measurements are possible for vessel walls with a thickness larger than 1.25 times the acquired voxel size. For thinner vessel walls, measurements were ambiguous and inaccurate. Because the vessel wall for all arteries of the circle of Willis is thinner than the acquired voxel size, early vessel wall thickening cannot be quantified from vessel wall images acquired with commonly used in vivo MR sequences.

Using the vessel wall image intensity allows for measurements for vessel walls thinner than the acquired voxel size (Chapter 4). However, such measurements are uncalibrated and qualitative, and are therefore only usable for visualizing relative differences in vessel wall thickness within a subject. We therefore proposed a neural network for measuring the subvoxel vessel wall thickness. Hereto, we imaged post-mortem circle of Willis specimens with a clinically-used sequence, and an ultra-high-resolution sequence. The neural network was trained to measure the vessel thickness from the clinically-used images. Thickness measurements on the ultra-high-resolution image, which were performed prior to training of the network, were used as reference for the network. Results showed that accurate subvoxel vessel wall thickness measurements are possible. In addition, we showed its feasibility on three in vivo vessel wall MR images of intracranial aneurysms.

In Chapter 5, we introduced software that allows for the measurement of the arterial diameter pulsatility and blood flow velocity pulsatility over the cardiac cycle, as measure for the arterial stiffness, from 4D phase contrast (velocity) MR (PC-MRI) images. Measurements on the arterial diameter pulsatility and blood flow velocity pulsatility using our software were validated against measurements from 2D PC-MRI for two locations of the internal carotid artery in 7 volunteers. Our software may potentially aid in identifying changes in arterial stiffness of the intracranial arteries caused by pathological changes to the vessel wall.

Lastly, we investigated the use of anomaly detection for the automated detection of brain pathology (Chapter 6). We evaluated the performance of our anomaly detection method on the detection of chronic brain infarcts. Besides the detection of brain infarcts, our proposed method also detected white matter hyperintensities, and image infarcts. Most interestingly, it also detected brain infarcts that were initially missed by the radiologist during radiological reading of the MR images. Anomaly detection is a powerful tool for the detection of brain pathology, and may aid diagnostic radiology by highlighting locations in the image that likely contain pathology.