Generative AI for quantitative MRI applications

. Medical imaging technologies are advancing rapidly, especially with the integration of Artificial Intelligence (AI)-based techniques. AI has significantly improved the optimization and interpretation of MR images. However, a major challenge is the lack of accessible data for building AI models. Even when data is available, it is often unsuitable for developing AI algorithms due to a lack of standardization. Generative models have been proposed to address this issue by learning the underlying probability distribution of the data. By using generative models, it is possible to optimize complex modalities like Magnetic Resonance Imaging (MRI), resulting in significantly improved image quality and accuracy. This project aims to explore novel generative model architectures that integrate relevant scan parameters to generate highly realistic datasets for training deep learning (DL) algorithms. The goal of this project is to propose a novel physics-based generative model for transferring data distributions in quantitative MRI (q-MRI) tasks, which typically require both MRI data and their corresponding parametric maps. The physics-based approach will generate data in a more intuitive and explainable manner than conventional generative methods. It also allows for indirect data transfer without privacy concerns and adapts to variations in scan protocols. Given that protocol mismatches are common in multi-center q-MRI studies, we expect this model to be a valuable tool for transferring data-driven knowledge between centers. We have recently demonstrated this approach for MR-based water-fat quantification in the liver. Therefore, the project will extend this concept to other q-MRI applications, such as T1, T2, and QSM mapping.