Simulation and Synthesis for Cardiac Magnetic Resonance Image Analysis
My PhD thesis, defended at Eindhoven University of Technology within the OpenGTN Marie Curie ITN-EID project under Prof. Dr. Ir. Marcel Breeuwer, develops methods for generating realistic cardiac MR images with corresponding ground-truth labels for supervised segmentation, addressing the scarcity of labelled cardiac MRI data. It connects two directions: physics-driven image simulation grounded in MR image-formation physics and the Bloch equations, published as a flexible framework for simulating cardiac MR images with variable anatomy and contrast, and data-driven image synthesis using conditional deep generative models, published as a label-conditional GAN framework and a VAE-based pathology synthesis method. Across both directions, the thesis shows that simulated and synthesized training data can match or improve the performance of deep-learning segmentation models trained on real data alone, particularly under limited or homogeneous training sets.
About this thesis
My PhD thesis is titled "Realistic Cardiovascular Magnetic Resonance Image Simulation and Synthesis for Medical Image Analysis," defended on 20 April 2023 at the Department of Biomedical Engineering, Eindhoven University of Technology, under supervisor Prof. Dr. Ir. Marcel Breeuwer, as part of the OpenGTN Marie Curie ITN-EID project.
The university's research portal does not publish a formal abstract for the thesis itself, so this description is built from the thesis's two connected, separately published research directions rather than a single quoted abstract.
Two connected directions
Physics-driven simulation. Grounded in MR image-formation physics and the Bloch equations, this direction produced a flexible framework for simulating cardiac MR images with variable anatomical and imaging characteristics, letting a large, diversified virtual population of labelled subjects be generated for downstream segmentation training. See Cardiac MRI Simulation Framework.
Data-driven synthesis. Using conditional deep generative models, this direction produced a mask-conditional GAN for translating segmentation labels into realistic cardiac images, and a VAE-based method for synthesizing 3D-consistent subjects with plausible heart pathologies. See Conditional Cardiac MR Image Synthesis and Cardiac Pathology Synthesis.
Why it matters
Across both directions, the thesis's consistent finding is that simulated and synthesized cardiac MR data, used either as a replacement for or augmentation to real data, can match or improve deep-learning segmentation performance, particularly when real labelled data is limited, homogeneous, or drawn from only one scanner vendor.