SU-E-I-19: Local Search Clustering Algorithm for DCE-MRI Analysis.

Area of Science:

• Medical Imaging
• Biophysics
• Computational Biology

Background:

• Dynamic contrasted enhanced (DCE)-MRI data often exhibit low contrast-to-noise ratio (CNR), particularly in the central nervous system, potentially leading to inaccurate analyses.
• Poor CNR can compromise the reliability of vascular permeability estimations derived from DCE-MRI.

Purpose of the Study:

• To develop and evaluate a local search clustering algorithm for analyzing DCE-MRI data.
• To improve the contrast-to-noise ratio (CNR) of DCE-MRI signals through clustering.
• To enhance the accuracy of vascular permeability determination from DCE-MRI data.

Main Methods:

• A local search clustering algorithm was developed to group proximal voxels with similar uptake curves and T1 values in DCE-MRI data.
• The algorithm iteratively expands clusters starting from seed voxels until a stopping criterion is met.
• The algorithm's effectiveness was tested using a 2D Shepp-Logan phantom with introduced variations in T1 values, permeability, and Gaussian noise, comparing results to voxel-by-voxel analysis.

Main Results:

• The local search clustering algorithm reduced errors in permeability parameter estimation by over 50% compared to traditional voxel-by-voxel analysis.
• The clustering approach demonstrated improved performance even when DCE-MRI data were corrupted with noise.
• The algorithm effectively segmented concentration time curves, enhancing apparent CNR within clustered regions.

Conclusions:

• The developed local search clustering algorithm effectively segments DCE-MRI data, improving signal quality.
• This method significantly enhances apparent CNR and reduces errors in vascular permeability estimation.
• The algorithm provides a robust approach for more accurate analysis of DCE-MRI data, especially in noisy conditions.