5T MRI and Metabolomics-Based Hemorrhage Risk Prediction in Moyamoya Disease

BACKGROUND MMD is a rare cerebrovascular disease characterized by stenosis or occlusion of the distal internal carotid artery bifurcation and the formation of collateral networks at the base of the brain. In China, the annual incidence is approximately 1.14 per 100,000 population and has been rising. MMD presents with ischemic or hemorrhagic manifestations; hemorrhagic-type MMD has an annual hemorrhage rate of approximately 17% and a mortality of 6.8%-20%, with a worse prognosis than the ischemic type. Autopsy and imaging studies indicate that dilatation and rupture of fragile, thin-walled collateral vessels are the principal cause of MMD hemorrhage. Hemodynamic changes and vessel wall alterations are associated with hemorrhage risk, but conventional 1.5T/3.0T MRA lacks the spatial resolution and signal-to-noise ratio required to quantify collateral vessels. 5.0T MRI offers substantially improved signal-to-noise ratio; its TOF-MRA quality is comparable to 7.0T and superior to 3.0T, enabling precise multidimensional assessment of collateral vessel morphology, blood flow, and vessel wall features. Meanwhile, metabolomic studies have consistently identified metabolic alterations (amino acid, lipid, purine/pyrimidine pathways) in patients with MMD, and the investigators' preliminary work has shown that glycine metabolism promotes angiogenesis and is associated with collateral vessel formation. However, the association between metabolomic alterations and the hemorrhagic subtype, and whether metabolic changes can serve as early biomarkers of hemorrhagic MMD, remains unknown.

STUDY DESIGN This is a single-center, prospective, observational cohort study conducted at Beijing Hospital. Patients with MMD or moyamoya syndrome confirmed by DSA who meet the eligibility criteria will be consecutively enrolled. Based on the initial presentation, participants will be classified into a hemorrhagic-type group and an ischemic-type group. All participants will undergo routine preoperative multimodal imaging (CT, CTA/P, MRA, DSA), a preoperative 5.0T MRI examination, and preoperative blood and urine sampling. Participants will be followed for 12 months; new-onset intracranial hemorrhage, analyzed per cerebral hemisphere, is the endpoint event, adjudicated by clinical symptoms and confirmatory head CT or MRI.

5T MRI ACQUISITION All MRI examinations will be performed on a 5T whole-body MR scanner (Siemens Healthcare, Erlangen, Germany) with a birdcage transmit and 32-channel receive head coil (Nova Medical, MA, USA). Sequences include T1W MPRAGE, optimized whole-brain isotropic TOF-MRA (0.40 mm isotropic resolution), an optimized single-slab TOF-MRA covering the middle cerebral artery and basal ganglia region (0.23×0.23×0.36 mm³), HR-VWI, and 3D ASL, together with conventional DWI/T2WI/SWI. Image quality will be graded before inclusion in the dataset.

IMAGING ANALYSIS A deep learning pipeline will be developed for quantitative collateral assessment: a VT-UNet 3D segmentation network will extract the vascular skeleton from TOF-MRA, followed by feature extraction with a Swin Transformer architecture. The skeleton will be converted to a tetrahedral mesh for computational hemodynamics, deriving wall shear stress (WSS), wall shear stress gradient (WSSG), oscillatory shear index (OSI), and relative residence time (RTT). Vessel wall features (stenosis site, remodeling pattern, stenosis degree, lumen area, wall thickness, remodeling index) will be manually annotated on HR-VWI by two independent readers (ICC > 0.8 required) and quantified with PyRadiomics (approximately 1,790 radiomic features). Brain perfusion and structural parameters (cerebral blood flow, cortical thickness, cortical surface area, brain and deep nuclei volumes) will be derived from 3D ASL and T1WI.

METABOLOMICS Preoperative venous blood (5 mL; plasma separated within 20 minutes) and urine (10 mL) will be collected and stored at -80 °C. High-throughput metabolomic profiling of these samples, including amino acid metabolites, will be performed to identify novel biomarkers of hemorrhagic MMD. STA specimens trimmed during direct bypass surgery will be formalin-fixed, paraffin-embedded, and analyzed by immunohistochemistry to cross-validate biomarkers and explore mechanisms.

MODEL DEVELOPMENT AND VALIDATION Clinical baseline data, 5T MRI features, hemodynamic parameters, and metabolomic biomarkers will be integrated. Feature selection and dimensionality reduction will use L1 regularization; baseline machine learning models (logistic regression, naive Bayes, SVM, neural network, KNN, decision tree) will be tested and, if necessary, ensemble learning methods will be applied, with hyperparameters tuned by grid search. Internal validation will use five repetitions of five-fold cross-validation; external validation will be performed in the validation cohort. The primary performance metric is the area under the receiver operating characteristic curve (AUROC), with accuracy, recall, precision, and F1 score as secondary metrics. The final model will be used to construct a hemorrhage risk stratification system.

SAMPLE SIZE Based on an annual hemorrhage rate of 17% and an estimated 5 predictor variables (10 events per variable), 50 hemorrhage events are required, yielding approximately 294 participants; adjusted for a 10% loss to follow-up and a 7:3 training/validation split, 500 participants will be enrolled.

SIGNIFICANCE This study will enable hemorrhage risk stratification of patients with MMD, early identification of high-risk individuals, and timely surgical revascularization, thereby reducing the disability and mortality of MMD and informing the comprehensive prevention and treatment of stroke.