Cerebrovascular Morphology as an Imaging Marker for Enhanced Stroke Patient Assessment

Abstract

Acute ischemic stroke (AIS) is a leading cause of death and disability with 15 million incidences and 5 million deaths annually, worldwide. Clinical AIS triage has to be both rapid and accurate to efficiently reestablish flow in the tissue-at-risk and improve patient outcomes, since, as is often said in acute stroke settings, ‘time is brain’. Response to treatment and clinical outcomes, however, vary significantly between patients since patient-specific cerebrovascular anatomy and collateral circulation, play a vital role in governing flow and reperfusion. This dissertation presents novel methods for automatic end-to-end stroke triage using image processing and machine learning techniques, based on cerebrovascular morphology, extracted from baseline imaging scans. This dissertation consists of three consecutive studies. In the first study, we developed and validated a method for accurate and fully automatic segmentation of the cranial vessel network from magnetic resonance angiography (MRA) or computed tomography angiography (CTA) scans and defined quantitative features of the vasculature. We combine a probabilistic vessel-enhancing filtering with an active-contour-based technique to segment angiograms and subsequently extract the vessel centerlines to calculate the geometrical properties of the vasculature, namely – total length, diameter, branching pattern, total volume, tortuosity and fractal dimension. Our method was extensively validated with image matching and registration performance metrics, using a biofidelic 3D phantom of the Circle-of-Willis region. It outperformed other commonly available segmentation tools and software. In the second study, we created a probabilistic atlas of human cerebrovasculature using MRA scans of 175 healthy adults and also extracted the corresponding geometric properties. These features were then used to study morphological changes in the brain vessels with normal aging, and the intersubject variations in the Circle of Willis (CoW) anatomy. We analyzed vascular alterations in 45 AIS and 50 Alzheimer’s disease (AD) patients – the two biggest cerebrovascular and neurodegenerative disorders – to characterize & quantify the underlying vascular pathology. We determined that the CoW is fully formed in only 35% of healthy adults and found significantly (p < 0.05) increased tortuosity and fractality, with increasing age and with disease in both AIS and AD. Along with the expected decrease in vascular volume in AIS, the AD cerebral vessels exhibited smaller diameter and more complex branching patterns, compared to age-matched healthy adults. These changes were heightened with AD progression from mild to moderate/severe dementia. In the final study, we use our validated algorithm to train and implement a CNN for the segmentation of 100 anonymized AIS patient MRA scans to obtain the 3D vessel networks. We then extract morphological features of the vessel tree and use these in conjunction with our cerebrovascular atlas to automatically detect the presence and location of the occlusion as well as quantitatively grade the collateral circulation development. Lastly, we use patient history, demographics and clinically assessed risk scores of stroke severity such as NIHSS and ASPECTS, in combination with our vascular features and automatically graded collateral index, to predict the patient-specific 90-day functional outcome on the modified Rankin scale (mRS) using machine learning. We compared the model’s performance against conventional outcome predictors and found that using vascular features significantly improved the prediction accuracy. The rapid, automatic process presented here is aimed at improving the accuracy and timing of detecting and localizing intracranial occlusions and the fidelity of predicting the functional outcomes in stroke patients using patient specific cerebrovascular morphology and imaging data. The overarching finding is that vascular morphology could potentially be used as a noninvasive imaging biomarker for neurologic disorders.