Real-Time, Inline Quantitative MRI Enabled by Scanner-Integrated Machine Learning: A Proof of Principle With NODDI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Advanced quantitative MRI (qMRI) techniques are often limited to research settings due to slow, offline parameter estimation.
• This hinders their clinical adoption and integration into routine workflows.

Purpose of the Study:

• To develop and validate a real-time, inline parameter estimation framework for advanced qMRI using neural networks.
• The goal is to enable 'clinical mode' qMRI, facilitating wider clinical use.

Main Methods:

• Customized Siemens Image Calculation Environment (ICE) to deploy trained neural networks (NNs) via ONNX Runtime.
• Trained two NNs offline using synthesized data from the neurite orientation dispersion and density imaging (NODDI) model.
• Demonstrated inline estimation in healthy volunteers and evaluated with synthetic data across two diffusion protocols.

Main Results:

• Successfully integrated NNs into ICE for inline, whole-brain NODDI parameter estimation in under 10 seconds.
• The workflow proved reproducible across different protocols, volunteers, and rescans.
• Exported DICOM parametric maps for further analysis, with NN estimates showing consistency with conventional fitting.

Conclusions:

• The real-time inline estimation framework overcomes a major barrier to clinical qMRI adoption.
• This generalizable approach facilitates efficient integration of advanced qMRI into clinical workflows.
• Future work includes incorporating pre-processing and evaluating in pathological conditions.