Cardiac Ultrasound Synthetic Data

M.A.Sc. thesis on conditional diffusion models for synthetic echocardiograms and segmentation data augmentation.

For my M.A.Sc. thesis at UBC, I developed ControlNet-based diffusion models to generate anatomically faithful synthetic echocardiograms. I evaluated the generated data through downstream segmentation, where augmenting U-Net models with synthetic images improved performance over real-only training, and explored CMMD as a metric aligned with downstream utility.

Focus: Biomedical Imaging · Ultrasound · Diffusion Models · Segmentation

Read the thesis in UBC’s cIRcle repository