Project Details
Projekt
Image-based personalized prediction of residual risk and prognosis of cardio/cerebrovascular disease
Applicants
Professorin Dr.-Ing. Anja Hennemuth; Professorin Dr. Jeanette-Esther Schulz-Menger; Dr. Kersten Villringer
Subject Area
Radiology
Cardiology, Angiology
Medical Informatics and Medical Bioinformatics
Cardiology, Angiology
Medical Informatics and Medical Bioinformatics
Term
since 2023
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 515294457
Individual risk prognosis for recurrent cardiovascular events and alteration of distant organs after acute cardio- and cerebrovascular events is a major unmet medical need, and improvements in this field may facilitate more stratified management strategies, improving patient outcomes. Advances in imaging technologies and in particular systematic analysis of imaging data using specific image features in correlation with clinical features provide the opportunity to personalize healthcare. This so-called radiomics approach aims to extract and use features, which describe shapes and spatial relations as well as texture, representing the underlying biology and pathophysiology. In this project, we aim to identify radiomics signatures in the brain and heart that can be used as biomarkers for an improved and personalized risk prediction in patients after an acute cardio- or cerebrovascular event. For this purpose, we will use cerebral and cardiac MRI data in combination with the comprehensive deep phenotyping- and long-term follow-up data available from participants of the Berlin Longterm Observation of Vascular Events (BeLOVE) cohort study. We will establish a web-based solution for bringing together clinical and imaging data, classification and supporting interactive analysis with imaging and clinical experts. In order to optimally consider anatomical and functional relations we intend to work with graph neural networks. Quantitative multiparametric maps of the brain and heart will be generated for different scenarios of structural alterations that is fibrosis of the heart and white matter lesions in the brain. In order to enable comparability for longitudinal data, a comprehensive quality assurance is carried out by analysing phantom as well as volunteer measurements. In a subsequent feedback loop with clinical experts, classifiers will be evaluated with regard to clinical interpretation and possible integration into clinical concepts. To enable interpretability, we will implement a visualization concept which considers the graph representations as well as the dependencies in the feature extraction and classification pipeline. An important milestone for integration of the risk prediction tool developed here into clinical practice is standardisation and optimization of the MR protocols in terms of clinically relevant classifiers in order to ensure a fast and targeted imaging and, as a result, a high acceptance by clinicians.
DFG Programme
Priority Programmes
