The proposed study will be based on a multimodal approach using 4D flow MRI, perfusion-weighted MRI (PWI), diffusion-weighted MRI (DWI) and high-resolution vessel wall imaging (VWI) together with patient information (demographics and clinical factors) to predict the risk of recurrent stroke of patients with intracranial atherosclerotic disease (ICAD) stenosis. This will allow integrating the vulnerability of the stenosis as well as the patient by assessing the hemodynamic impact, plaque stability, and stroke lesion pattern together with patient information into a prediction model. PWI will provide tissue perfusion, VWI will provide plaque stability, DWI will provide stroke lesion pattern and 4D flow MRI will provide macroscopic hemodynamics of the circle of Willis (CoW). We will concentrate on the following innovative developments: 4D flow MRI: In order to allow 4D flow MRI scanning with a high dynamic velocity range (necessary to measure slow and fast velocities simultaneously), we recently developed dual-venc 4D flow MRI. However, this method suffers from extended scan tome of an already long acquisition. We, therefore, aim to minimize scan time for dual-venc 4D flow MRI scan while using the required spatial resolution and volume coverage, targeting 5-10 minutes so that this sequence can be added to clinical protocols.
We aim to achieve this by integrating compressed sensing acceleration. Rigorous testing of the sequence will be done in phantom experiments as well as in a healthy test-retest control study. Data Analysis and Outcome Prediction: Currently, the multi-modal information that can be acquired with MRI has not been combined and used for comprehensive prediction of recurrent stroke risk in ICAD. Information that can be acquired from different MRI modalities may be critical in characterizing ICAD patient status. We will develop a new analysis tool that combines all data into a single network graph. All imaging data will be reported relative to supplying the intracranial artery of the CoW by using the vascular territory region of interest analysis. This will allow gathering all imaging parameters in a network graph. In a cross-sectional patient study, we will use combined data to see if it enables differentiation between healthy subjects, ICAD subgroups. Patient Study:
In Aim 3, we will develop a machine-learning algorithm to predict which of the patients are at risk of experiencing a recurrent stroke. In order to achieve this, we will enroll a total of 150 ICAD patients from two institutions (Northwestern Memorial Hospital and San Francisco General Hospital). The combined data from the four different MR modalities and all other patient information will be used to identify only the discriminative features. This will be realized by using support vector machine recursive feature elimination to rank features associated with the risk of an ischemic event. The SVM will be trained and tested using information from the patient's clinical follow-up as outcome measure. The outcome (ischemic event or death yes/no)) will enable the development of the SVM classifier to predict outcome.
(2 ? 3 sentences) In the proposed study, we will develop an advanced intracranial MRI protocol together with a comprehensive image processing pipeline allowing full characterization of intracranial atherosclerotic disease (ICAD). Imaging biomarkers together with clinical and demographic factors will enable the identification of patients at risk to suffer recurring stroke. Machine learning methods will be used to predict risk. Successful completion will allow for better identification of patients who could benefit from endovascular stenting to reduce stroke recurrence.
