Learning the Regularization in DCE-MR Image Reconstruction for Functional Imaging of Kidneys

Area of Science:

• Medical Imaging
• Radiology
• Biomedical Engineering

Background:

• Dynamic contrast-enhanced MRI (DCE-MRI) is crucial for assessing kidney anatomy and function via tracer kinetic (TK) modeling.
• High temporal resolution is needed for accurate arterial input function (AIF) measurement in DCE-MRI, but accelerated imaging introduces under-sampling artifacts.
• Traditional compressed sensing (CS) methods reduce artifacts but can over-smooth signals, compromising functional parameter accuracy.

Purpose of the Study:

• To develop a novel deep learning approach for artifact reduction in kidney DCE-MRI.
• To improve the accuracy of quantitative functional imaging markers derived from DCE-MRI.
• To overcome the limitations of conventional CS reconstruction in balancing artifact removal and signal preservation.

Main Methods:

• Proposed a single-image trained deep neural network (DNN) for artifact reduction.
• Implemented regularization by generating images from a lower-dimensional representation, avoiding traditional penalty terms.
• Compared the DNN approach against CS reconstructions with varying regularization weights.

Main Results:

• The DNN approach effectively reduced under-sampling artifacts in reconstructed DCE-MRI images.
• Kidney biomarkers estimated using the DNN method showed high correlation with ground truth markers.
• The proposed method preserved the accuracy of functional imaging markers, unlike heavily regularized CS methods.

Conclusions:

• The proposed deep learning method offers a promising alternative for artifact reduction in kidney DCE-MRI.
• This approach enhances the accuracy of quantitative kidney function assessment without sacrificing image quality.
• It provides a superior balance between artifact suppression and preservation of functional information compared to standard CS techniques.