Multispectral MRI segmentation of age related white matter changes using a cascade of support vector machines

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Machine Learning

Background:

• White matter changes (WMC) are associated with cognitive impairment and depression in the elderly.
• Manual segmentation of WMC is labor-intensive and susceptible to subjective bias, hindering research.
• Developing automated methods is crucial for efficient and objective WMC analysis.

Purpose of the Study:

• To present a fast, fully automated method for white matter changes (WMC) segmentation.
• To overcome the limitations of manual outlining procedures in WMC research.
• To develop a robust and efficient algorithm for WMC segmentation using machine learning.

Main Methods:

• A cascade of reduced support vector machines (SVMs) with active learning was employed for WMC segmentation.
• The framework included pre-processing, classification (training and core segmentation), and post-processing steps.
• T1-weighted and FLAIR MRI sequences from 102 subjects were utilized, with manual WMC outlines serving as the reference standard.

Main Results:

• The automated segmentation framework achieved high accuracy, with a sensitivity of 90% and specificity of 99.5%.
• The receiver operating curve (ROC) technique confirmed the effectiveness and robustness of the classification.
• The method demonstrated competitive performance compared to manual outlining, significantly reducing processing time and subjectivity.

Conclusions:

• A fully automated and efficient algorithm for WMC segmentation has been developed.
• The proposed SVM-based framework offers a robust and objective alternative to manual segmentation methods.
• This approach is adaptable to different MRI sequences without requiring modifications to the core algorithm.