Classification of LV wall motion in cardiac MRI using kernel Dictionary Learning with a parametric approach

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Biomedical Engineering

Background:

• Assessing Left Ventricle (LV) function is crucial for diagnosing cardiac diseases.
• Cardiac cine-Magnetic Resonance Imaging (MRI) provides detailed anatomical and functional information.
• Quantifying LV wall motion accurately remains a challenge in cardiac MRI analysis.

Purpose of the Study:

• To propose a novel parametric approach for assessing LV wall motion in cardiac cine-MRI.
• To develop a method for extracting dynamic functional information from LV contraction.
• To improve the accuracy of LV function classification using advanced machine learning techniques.

Main Methods:

• Extracted Spatio-temporal image profiles and Time-Signal Intensity Curves (TSICs) from cardiac MRI sequences.
• Constructed parameters from TSICs, including average curves (clustering), curve skewness, and cross-correlation.
• Trained a sparse classifier based on kernel Dictionary Learning (DL) using these parameters.
• Compared DL performance against Support Vector Machine (SVM) and Discriminative Dictionary Learning.

Main Results:

• The parametric approach effectively captured dynamic LV contraction information.
• Classification performance was evaluated using parameters derived from average curves and skewness.
• Kernel DL with a radial basis function kernel achieved the highest accuracy of approximately 94%.

Conclusions:

• The proposed parametric method offers a robust approach for LV wall motion assessment in cardiac MRI.
• Kernel Dictionary Learning demonstrates superior performance for classifying LV function based on extracted parameters.
• This technique holds potential for improved clinical diagnosis and patient management in cardiology.