Magnetic Resonance Imaging is a powerful and non-invasive diagnostic tool burdened by long scan time leading to increased costs and patient discomfort. Over the decades, several signal processing techniques to recover artifact-free images from sub-Nyquist sampled data (parallel imaging, compressed sensing) have been employed to drastically reduce the examination time. In particular, the recent growth in popularity of artificial intelligence in all the fields of science led to the development of machine learning approaches to MRI image reconstruction, allowing further acceleration. In this talk, we will review the main machine learning -based image reconstruction technique (i.e., image-/kspace-/cross-domain; data-driven/model-based) and discuss their advantages and pitfalls. Finally, we will briefly introduce DeepMR, an MR image reconstruction framework developed within the INFN-CSN5 PREDATOR project focused on the application of ML-powered techniques to quantitative MR.
