Support vector machines in DSC-based glioma imaging: suggestions for optimal characterization
Frank G Zöllner1, Kyrre E Emblem, Lothar R Schad
1Computer Assisted Clinical Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany. frank.zoellner@medma.uni-heidelberg.de
Magnetic Resonance in Medicine
|June 22, 2010
Summary
Support vector machines (SVMs) effectively characterize gliomas using dynamic susceptibility contrast magnetic resonance perfusion imaging (DSC-MRI) data. This automated approach aids in presurgical glioma grading, improving diagnostic accuracy.
Area of Science:
- Neuroimaging
- Machine Learning
- Oncology
Background:
- Dynamic susceptibility contrast magnetic resonance perfusion imaging (DSC-MRI) is crucial for glioma characterization.
- Support vector machines (SVMs) offer a method for prospective patient characterization based on prior data.
Purpose of the Study:
- To compare four different SVM models for glioma characterization using DSC-MRI data.
- To evaluate the feasibility of combining automated tumor segmentation with SVM classification for presurgical glioma assessment.
Main Methods:
- Features were extracted from automatically segmented tumor volumes across 101 DSC-MR examinations.
- Four distinct SVM models were trained and compared, with data rebalancing to ensure equal class representation.
- Model performance was assessed using an independent test dataset.
Main Results:
- All SVM models demonstrated high prediction accuracies exceeding 82% after data rebalancing.
- A SVM model utilizing a radial basis function kernel achieved the best discrimination.
- The best model correctly predicted low-grade gliomas with 83% accuracy and high-grade gliomas with 91% accuracy.
Conclusions:
- Automated tumor segmentation combined with SVM classification is a feasible and powerful tool for presurgical glioma characterization.
- This integrated approach enhances diagnostic capabilities, aiding in treatment planning.
