Cancer patients can greatly benefit from personalized medicine, in which quantitative magnetic resonance imaging (MRI) biomarkers are used to select the optimal personalised treatment. Particularly, intravoxel incoherent motion MRI (IVIM) and dynamic contrast-enhanced MRI (DCE) provide two key biomarkers: tissue perfusion and diffusion. However, IVIM and DCE images are often of poor quality, low resolution, inaccurate, imprecise, and, therefore, hard to interpret. Consequently, despite their great potential, these quantitative MRI (qMRI) techniques are implemented only in a limited number of expert clinics.
We will fundamentally change how qMRI is obtained and develop innovative explainable unsupervised physics-informed AI approaches for generating qMRI images that are of clinical quality. This will open the way to personalised treatments.
These AI-driven frameworks will be tailored to suit the specific needs of qMRI. We will train networks to obtain accelerated high-resolution qMRI parameter maps, without the need for ground truth references. We will develop explainable AI and produce uncertainty estimates alongside the parameter maps. Such maps are currently unavailable. Finally, we will test how accurately responding versus non-responding head-and-neck cancer patients receiving radiotherapy can be distinguished.
People on the project:
