Greybox: A hybrid algorithm for direct estimation of tracer kinetic parameters from undersampled DCE-MRI data

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biophysics

Background:

• Dynamic Contrast-Enhanced (DCE) MRI is crucial for predicting Tracer Kinetic (TK) parameters.
• Existing deep learning and iterative methods have limitations in handling undersampled DCE-MRI data.
• A need exists for robust algorithms that can accurately estimate TK parameters from undersampled data.

Purpose of the Study:

• To introduce Greybox, a hybrid algorithm combining model-based and deep learning (DL) approaches.
• To address the non-linear inverse problem of directly estimating TK parameters from undersampled DCE-MRI data.
• To develop an algorithm invariant to varying undersampling rates.

Main Methods:

• Inspired by plug-and-play algorithms, Greybox uses a UNET-based DL prior for TK parameter estimation.
• An iterative gradient-based optimization scheme is employed with the DL prior.
• The method was validated on brain, breast, and prostate DCE-MRI datasets with varying undersampling rates (8x, 12x, 20x) using Radial Golden Angle (RGA) undersampling.

Main Results:

• Greybox demonstrated superior performance compared to direct reconstruction methods.
• The algorithm achieved up to a 3 dB improvement in peak signal-to-noise ratio (PSNR) for estimated TK parameters.
• Statistical tests confirmed the significant performance enhancement.

Conclusions:

• The Greybox hybrid algorithm offers state-of-the-art performance for DCE-MRI reconstruction.
• It is the first method to integrate DL-based priors within plug-and-play frameworks for improved DCE-MRI analysis.
• Greybox effectively solves the non-linear inverse problem in DCE-MRI, even with undersampled data.