A Novel Statistical Optimization Algorithm for Estimating Perfusion Curves in Susceptibility Contrast-Enhanced MRI.

Area of Science:

• Medical Imaging
• Neuroscience
• Biophysics

Background:

• Dynamic susceptibility contrast-enhanced magnetic resonance imaging (DSC-MRI) is crucial for assessing intravascular indicator dynamics in brain research.
• Conventional curve-fitting methods like nonlinear optimization and logarithmic linear least-squares (LL-LS) have limitations, including computational expense, initial value dependency, and degraded performance with limited or outlier data.

Purpose of the Study:

• To develop a more robust and efficient method for analyzing concentration-time curves in DSC-MRI.
• To improve the assessment of brain physiology and pathophysiology by overcoming limitations of existing curve-fitting techniques.

Main Methods:

• Transformed the DSC-MRI perfusion curve-fitting problem into a gamma-distribution-fitting problem.
• Treated concentration-time curves as a random sample from a gamma distribution.
• Applied a robust maximum-likelihood estimation (MLE) algorithm to solve the transformed problem.

Main Results:

• The proposed MLE method demonstrated significantly superior performance compared to nonlinear Levenberg-Marquardt (L-M) and LL-LS methods.
• The new approach maintained the advantages of LL-LS, including ease of implementation, low computational load, and no requirement for initial values.
• Performance was validated using both synthetic and real DSC-MRI data.

Conclusions:

• The proposed gamma-distribution-fitting approach using MLE is a highly attractive alternative for analyzing intravascular indicator dynamics in clinical DSC-MRI applications.
• The study provides practical recommendations for optimizing traditional LL-LS and L-M methods for more reliable estimations.