EE Distinguished Lecturer Seminar: Inverse problems in medical imaging – view the invisible
Abstract: Over the last decade, thanks to the development of both mathematical methods and computational power, medical imaging has been evolving rapidly from classical image visualization and post-processing to the extraction of more subtle quantitative information about the organ/tissue at hand. This domain is now known as computational medical imaging. In this context, I will present how the resolution of a regularized linear inverse problem can reveal unprecedented information in medical imaging. I will first consider the relatively recent modality of diffusion Magnetic Resonance Imaging (MR), and show how we can extract unique information from the raw data by properly modeling the forward problem and carefully inverting it. Our work allows extracting information about the brain tissue architecture that is several orders of magnitude smaller than the MR imaging resolution, leading to so-called brain microstructure analysis. I will also show how different MR contrasts can be used jointly to reconstruct an even more complete picture of the tissue microstructure, paving the way towards what I call MR microscopy. I will also show how the same concepts apply to other medical imaging problems, namely ultrasound image reconstruction and radiotherapy planning, and I will finish by discussing how machine learning can contribute to this field.