Joint Denoising of Measurements and Reconstructions in Ultrasound Imaging – Can Deniz Bezek

Medical imaging aims to recover underlying tissue properties from inaccurate and incomplete measurements, often relying on simplified imaging models. Analytical reconstruction methods employ hand-crafted regularization and are sensitive to noise assumptions and parameter tuning. Among deep learning solutions, plug-and-play (PnP) methods enable learned regularization while incorporating imaging physics during inference, outperforming purely data-driven methods. In this seminar, I will present a framework that jointly denoises both the input measurements and the resulting reconstructions in their respective domains. The proposed framework consists of a refinement network that corrects degraded measurements while compensating for imaging model simplifications, combined with a diffusion-based PnP reconstruction method. I will demonstrate the approach on speed-of-sound imaging, a particularly challenging problem in quantitative ultrasound image reconstruction.

About Can Deniz Bezek