Real-time quantitative MRI enabled by scanner integrated machine learning: a proof of principle with NODDI

Motivation: Resource intensive offline image post-processing and parameter estimation inhibit the clinical integration and feasibility of quantitative MRI (qMRI). Goal(s): To achieve real-time, online qMRI parameter estimation with a trained neural network (NN) integrated into a vendor's image reconstruction environment, facilitating and encouraging clinical adoption of qMRI. Approach: A NN for estimating parameters of the NODDI diffusion MRI model was trained offline using synthetic data. Its trained state was exported into ONNX format (a cross-language ML framework) and integrated into a custom Siemens Image Calculation Environment (ICE) module. Results: The integrated NN performs accurate, online, real-time NODDI parameter estimation of in vivo data. Impact: Real-time, online qMRI parameter estimation with the presented strategy resolves a key practical barrier to clinical uptake of qMRI methods and enables their efficient integration into clinical workflows.