Cardiac Spatio-Temporal Volumetric Mesh Reconstruction for Local Strain Computation and Monitoring Local Treatment

• Applicant: Dr. Anne-Marie Rickmann
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing ; Methods in Artificial Intelligence and Machine Learning ; Radiology
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 559051914

Left ventricular (LV) remodeling following acute myocardial infarction (MI) is a critical concern in the treatment of ischemic heart disease, a leading cause of mortality worldwide. Clinically, abnormalities in wall motion, such as dyskinesis or akinesis, are used to detect myocardial injury, typically assessed through visual analysis of heart movement. Myocardial strain analysis, a computational method that quantifies myocardial deformation, serves as an early and sensitive biomarker for detecting myocardial disease. Current research on myocardial strain quantification has primarily followed two perspectives: the biomechanical engineering approach and the image analysis approach. The biomechanical perspective relies on finite element (FE) models of the heart, developed based on material properties and boundary conditions to simulate heart motion. While these models are sophisticated, they often lack direct validation against patient-specific imaging data, leading to potential discrepancies between the simulated and actual heart movements. On the other hand, the image analysis perspective focuses on extracting displacement data directly from imaging modalities such as echocardiography (Echo) or cardiac computed tomography (CT) to compute strain. This approach ensures that the displacements align with observed heart movements, but it may not account for the physical realism required for accurate simulations of heart behavior. Previous research in image-based mesh reconstruction has generally focused on surface meshes or isolated time frames, often neglecting the integration of physical priors that would enhance the physiological relevance of the computed displacements. Despite advancements in both fields, a significant gap remains in combining biomechanical and image-based approaches. Existing methods often fail to utilize the full temporal sequence of cardiac imaging and do not adequately incorporate essential physical priors, resulting in models that may not accurately reflect the true physiological conditions. Moreover, the predominant use of surface-based methods limits the ability to capture the full scope of myocardial deformation. To address these challenges, my research proposes the development of a deep learning model, that learns spatio-temporal volumetric meshes from cardiac image sequences, integrating biomechanical priors. This model will be trained across different imaging modalities, capturing the entire cardiac sequence and ensuring that the displacements are both image-accurate and biomechanically plausible.

• DFG Programme: WBP Fellowship
• International Connection: USA
• Host: Professor James Duncan, Ph.D.