Medizinische Universität Graz Austria/Österreich - Forschungsportal - Medical University of Graz
SHR
Neuro
Krebs
Kardio
Stoffw
Microb
Lipid
Bild-basiertes Lernen für prädiktive personalisierte Modelle der Herzfunktion
- Abstract
- The heart is an electrically controlled mechanical pump, which drives blood flow from its cavities into the vascular system through deformation of its walls. Its function emerges from a bidirectional transduction of processes across multiple physical systems as well as spatial and temporal scales. This multiscale-multiphysics nature renders the heart an inherently challenging organ to study through reductionist approaches alone. This fundamental problem can be addressed by using biophysically detailed in silico models. Today in basic sciences such models are considered an indispensable element in any advanced physiology study as they offer the unique ability to mechanistically analyze cause-effect relationships at high spatio-temporal resolutions in the intact organ across scales and physics.
Driven by recent advances in medical imaging and simulation technologies, a translational trend emerged which is geared towards evolving modeling into a clinical modality. In the clinic image-based analysis of the dynamics of electrophysiological activity, deformation and blood flow is of pivotal importance in the diagnostic assessment of cardiac function. Due to the sparse and noisy nature of medical image data computational analysis tools are essential for providing a more objective quantitative evaluation of a patient's condition. Moreover, relevant biomarkers cannot be measured directly and must be derived from images by computational analysis, a process which is often enriched by PDE models of the underlying physics. To further gain clinical traction such image-informed in silico models must be personalized for a given patient to provide clinically useful biomarkers, which can serve to indicate or stratify disease or allow for virtual testing of treatments and prediction of outcomes acutely and longitudinally. Personalization requires the conception of sophisticated data assimilation procedures to facilitate model adaptation to a patient's cardiac anatomy and function. Such data-model fusion techniques are highly non-trivial to devise for clinical electro-mechano-fluidic (EMF) models due to the large number of and the non-linear relation between model components that must be fit.
This project combines expertise in cardiac modeling and simulation, computational image analysis, learning - based data assimilation techniques and high performance computing algorithms to render feasible the development of robust and efficient data-model fusion framework for the image-based personalization of models of total heart function. The long term aim is the development of a clinical assist modality for the optimization and outcome prediction of valve (aortic/mitral) replacement therapies in the left heart. It is anticipated that such personalized EMF models will play an essential role in informing clinical decision making in future precision therapies.
- Schlagworte
- Computer-assistierte Kardiologie
- Hochleistungsrechnen
- Maschinenlernen
- Mathematische Optimierung
- Medizinische Bildverarbeitung
- Projektleitung:
- Plank Gernot
- Laufzeit:
- 01.01.2017-31.12.2020
- Subprogramm
- BioTechMed Flagship Project
- Art der Forschung
- Grundlagenforschung
- Mitarbeiter*innen
- Plank G., Projektleiter*in
- Neic A., Projektmitarbeiter*in
- Karabelas E., Projektmitarbeiter*in
- Gsell M., Projektmitarbeiter*in
- Gillette K., Projektmitarbeiter*in
- Beteiligte MUG-Organisationseinheiten
- Lehrstuhl für Medizinische Physik und Biophysik
- Projektpartner
- Institut für maschinelles Sehen und Darstellen, Österreich
- Karl Franzens Universität Graz, Institut für Mathematik und wissenschaftliches Rechnen, Österreich
- Gefördert durch
- BioTechMed, Österreich
- Publizierte Projektergebnisse
- > A Framework for the generation of digital twins of... Med Image Anal. 2021; 71:102080
- > Using machine learning to identify local cellular ... Europace. 2021; 23(Supplement_1):i12-i20-i12-i20
- > GEASI: Geodesic-based earliest activation sites id... Int J Numer Method Biomed Eng. 2021; 37(8):e3505
- > An Inverse Eikonal Method for Identifying Ventricu... J Comput Phys. 2020; 419:
- > Inverse localization of earliest cardiac activatio... J Math Biol. 2019; 79(6-7): 2033-2068.
- > Towards a Computational Framework for Modeling the... Front Physiol. 2018; 9(12):538-538