Deconvolution of dynamic contrast-enhanced MRI data by linear inversion: choice of the regularization parameter

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging
• Signal Processing

Background:

• Dynamic contrast-enhanced MRI (DCE-MRI) deconvolution is crucial for assessing tissue perfusion.
• Truncated Singular Value Decomposition (TSVD) is a common deconvolution technique.
• Selecting the optimal truncation threshold in TSVD is challenging and impacts results.

Purpose of the Study:

• To optimize and compare two pixel-by-pixel threshold selection methods for TSVD in DCE-MRI: Generalized Cross Validation (GCV) and L-curve Criterion (LCC).
• To evaluate the impact of these methods on the accuracy of perfusion parameter estimation and image quality.

Main Methods:

• Implementation and optimization of GCV and LCC for TSVD threshold selection on a per-pixel basis.
• Comparison of optimized GCV and LCC against existing methods using simulated DCE-MRI data.
• Assessment of the residue function, maximum residue, signal-to-noise ratio (SNR) convergence, and estimation of perfusion parameters like mean transit time.

Main Results:

• Both optimized GCV and LCC significantly improve the estimation of the residue function and its maximum.
• Perfusion is more accurately estimated, especially at small mean transit times, with vanishing oscillations in the solution.
• Improved image contrast and increased sensitivity to perfusion abnormalities were observed, with GCV being computationally faster than LCC.

Conclusions:

• Optimized GCV and LCC provide robust and improved deconvolution for DCE-MRI compared to conventional methods.
• These methods enhance the accuracy of perfusion quantification and diagnostic image quality.
• GCV is recommended for its comparable performance and faster calculation time, though potential instabilities require further optimization.