Related Experiment Videos
Predictive modeling in glioma grading from MR perfusion images using support vector machines
Kyrre E Emblem1, Frank G Zoellner, Bjorn Tennoe
1Department of Medical Physics, Rikshospitalet University Hospital, Oslo, Norway. kyrre.eeg.emblem@rikshospitalet.no
Magnetic Resonance in Medicine
|September 26, 2008
Summary
This study introduces a predictive support vector machine (SVM) model for glioma grading using MR perfusion imaging. The model
Area of Science:
- Neuroimaging
- Oncology
- Machine Learning
Background:
- Glioma grading is crucial for treatment planning and prognosis.
- Magnetic Resonance Imaging (MRI) perfusion techniques offer quantitative biomarkers for tumor characterization.
- Predictive modeling for glioma grading using MR perfusion data remains underexplored.
Purpose of the Study:
- To implement a support vector machine (SVM) model for glioma grading.
- To utilize tumor blood volume histogram signatures from MR perfusion images as input features.
- To evaluate the diagnostic accuracy and sample size sensitivity of the predictive SVM model.
Main Methods:
- Dynamic Susceptibility Contrast (DSC) MRI was performed on 86 patients with gliomas.
- Histogram signatures from 53 patients were used to train a predictive SVM model.
- The SVM model was validated on the remaining 33 patients.
Main Results:
- The SVM model achieved a true positive rate (TPR) of 0.76 and a true negative rate (TNR) of 0.82 on the validation set.
- Model performance, including interobserver agreement and TPR, significantly improved with an increasing training sample size (P < 0.001).
- The model demonstrated diagnostic utility in differentiating glioma grades.
Conclusions:
- A predictive SVM model using MR perfusion histogram signatures shows promise for glioma grading.
- The model's accuracy and reliability are enhanced by larger training datasets.
- This approach can potentially aid clinicians in diagnosing glioma grade non-invasively.