MIBE Online Seminar: Physical Constraints-Aware Machine Learning for Neurovascular Image Computing

Titel: Physical Constraints-Aware Machine Learning for Neurovascular Image Computing

Sprecher: Suprosanna Shit, GCB doctoral candidate at the Chair for Image Based Biomedical Modeling, IBBM (Menze)

Datum und Uhrzeit: Dienstag, 07. Dezember, 13:00 Uhr

Abstract: 
Neurovascular disease is a leading cause of adult disabilities and death. Broadly, neurovascular disease refers to any abnormality of the blood vessels within or supplying blood to the brain and spine. This includes stenosis, i.e., narrowing of the arteries, which decreases blood flow to the brain and increases the risk of stroke. Additionally, a neurovascular disease may cause a weakening of the arteries that can lead to brain aneurysms and other conditions that increase the risk of a stroke. Medical doctors rely on local quantitative characterization of blood flow dynamics for clinical intervention and therapeutic planning in neurovascular diseases. 3D phase-contrast MRI (4D flow MRI) is an imaging modality that allows us to acquire a point-wise,in-vivo measurement of the instantaneous velocity of blood flow without administering intravenous contrast material. Thus acquired velocity fields present a valuable image basis to derive non-invasive hemodynamic biomarkers. Further, segmentation of the vascular network is a constant preprocessing task for the 4D flow analysis. Thus, I will focus on three major topics: 1) I will discuss a novel topology-preserving vessel segmentation framework. 2) Next, I will present a novel new neural network-based fast pressure inference scheme and the validation study using computational fluid dynamics (CFD) simulated data on patient-specific blood vessel geometries. 3) Lastly, I will present a fast and accurate neural network-based spatiotemporal super-resolution of 3D time-varying velocity fields using in-vivo and CFD simulated data of intracranial aneurysm.